[go: up one dir, main page]

WO2021146927A1 - Method and apparatus for sensor fault detection - Google Patents

Method and apparatus for sensor fault detection Download PDF

Info

Publication number
WO2021146927A1
WO2021146927A1 PCT/CN2020/073539 CN2020073539W WO2021146927A1 WO 2021146927 A1 WO2021146927 A1 WO 2021146927A1 CN 2020073539 W CN2020073539 W CN 2020073539W WO 2021146927 A1 WO2021146927 A1 WO 2021146927A1
Authority
WO
WIPO (PCT)
Prior art keywords
sensor
data
time period
during
fault
Prior art date
Application number
PCT/CN2020/073539
Other languages
French (fr)
Inventor
Xiao Liang
Daniel Schneegass
Peng Wei Tian
Yi Mao
Original Assignee
Siemens Schweiz Ag
Siemens Ltd., China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Siemens Schweiz Ag, Siemens Ltd., China filed Critical Siemens Schweiz Ag
Priority to PCT/CN2020/073539 priority Critical patent/WO2021146927A1/en
Publication of WO2021146927A1 publication Critical patent/WO2021146927A1/en

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01DMEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D3/00Indicating or recording apparatus with provision for the special purposes referred to in the subgroups
    • G01D3/08Indicating or recording apparatus with provision for the special purposes referred to in the subgroups with provision for safeguarding the apparatus, e.g. against abnormal operation, against breakdown
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0004Gaseous mixtures, e.g. polluted air
    • G01N33/0009General constructional details of gas analysers, e.g. portable test equipment
    • G01N33/0027General constructional details of gas analysers, e.g. portable test equipment concerning the detector
    • G01N33/0036General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
    • G01N33/0037NOx
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0004Gaseous mixtures, e.g. polluted air
    • G01N33/0009General constructional details of gas analysers, e.g. portable test equipment
    • G01N33/0027General constructional details of gas analysers, e.g. portable test equipment concerning the detector
    • G01N33/0036General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
    • G01N33/0039O3
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0004Gaseous mixtures, e.g. polluted air
    • G01N33/0009General constructional details of gas analysers, e.g. portable test equipment
    • G01N33/0027General constructional details of gas analysers, e.g. portable test equipment concerning the detector
    • G01N33/0036General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
    • G01N33/004CO or CO2
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0004Gaseous mixtures, e.g. polluted air
    • G01N33/0009General constructional details of gas analysers, e.g. portable test equipment
    • G01N33/0027General constructional details of gas analysers, e.g. portable test equipment concerning the detector
    • G01N33/0036General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
    • G01N33/0042SO2 or SO3
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0004Gaseous mixtures, e.g. polluted air
    • G01N33/0009General constructional details of gas analysers, e.g. portable test equipment
    • G01N33/007Arrangements to check the analyser
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01WMETEOROLOGY
    • G01W1/00Meteorology
    • G01W1/02Instruments for indicating weather conditions by measuring two or more variables, e.g. humidity, pressure, temperature, cloud cover or wind speed
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02ATECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A50/00TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE in human health protection, e.g. against extreme weather
    • Y02A50/20Air quality improvement or preservation, e.g. vehicle emission control or emission reduction by using catalytic converters

Definitions

  • the present invention relates to techniques of sensor, and more particularly to a method, apparatus and computer-readable storage medium for sensor fault detection.
  • a sensor/transducer is a device that responds to a physical stimulus (such as heat, light, sound, pressure, magnetism, or particular motion) and transmits a resulting impulse (as for measurement or control) , which we usually call as sensor data.
  • a physical stimulus such as heat, light, sound, pressure, magnetism, or particular motion
  • a resulting impulse as for measurement or control
  • sensors are deployed near monitored devices. Due to environmental factors, sensors may not work well, for example, being insensitive, reporting wrong data value, etc. unfortunately, such malfunctions are usually not easy to detect in comparison to no data transmission or blackout.
  • Embodiments of the present disclosure include methods, apparatuses for sensor fault detection, which can detect above sensor malfunctions timely and reliably.
  • a method, apparatus, system, computer-readable medium for sensor fault detection are presented.
  • Sensor data are collected and similarity of sensor data from different sensors are compared regardless of influence of event happened affecting sensor data and sensor location differences. Whether a sensor is in fault is based on the calculated similarity.
  • malfunctions such as insensitiveness, wrong data value reporting, etc. can be detected easily.
  • the detection result can be accurate and faults can be detected in time.
  • sensor data can be collected during a first historical time period and the sensor can be determined whether to be in fault during the first historical time period.
  • Each calculated similarity can be compared with a pre-defined threshold for the pair of the first sensor and the corresponding second sensor during the first historical time period, and whether the first sensor is in fault can be determined during the first historical time period based on each comparison result.
  • the threshold can be determined as such: sensor data of the first sensor and of the at least one second sensor can be collected during at least two second historical time periods; and for each second sensor, similarity of the collected sensor data of the first sensor with the second sensor can be calculated for each second historical time period during the corresponding second historical time period; then the pre-defined threshold for the pair of the first sensor and the second sensor can be determined based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
  • the probability of historical value can be considered instead of the exact value, which avoids using the average value as the threshold , weakens the influence of historical absolute value on the future trend, and makes the threshold setting more accurate and can reflect the change of time.
  • a method, apparatus, system, computer-readable medium for sensor fault detection are presented.
  • a model is trained based on historical sensor data, historical forecast data affecting sensor data and historical observed data affecting sensor data.
  • Sensor data can be forecast based on a model, and the forecast sensor data and real time sensor data collected can be compared for determination on whether the sensor is in fault.
  • the model used for sensor data forecast is trained based on the historical forecast data and time factors, which can handle the uncertainty of forecast data well.
  • FIG. 1 depicts a block diagram of a system for sensor fault detection.
  • FIG. 2A and FIG. 2B depict similarity of sensor under normal status and fault status.
  • FIG. 3 depicts a flow diagram of a method for sensor fault detection in accordance with one embodiment of the present disclosure.
  • FIG. 4 depicts a block diagram of a system for sensor fault detection.
  • FIG. 5 depicts structure of a model for forecasting sensor data.
  • FIG. 6 depicts a flow diagram of a method for sensor fault detection in accordance with one embodiment of the present disclosure.
  • the articles “a” , “an” , “the” and “said” are intended to mean that there are one or more of the elements.
  • the terms “comprising” , “including” and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements.
  • FIG. 1 depicts a block diagrams of a system 100 in accordance with one embodiment of the present disclosure. As shown in FIG. 1, there are sensors located on different places in an area, such as a city, a district in a city, a community or a building.
  • these sensors are used for monitoring air quality of a city, there are sensor data 31, 32 and other input data 33.
  • the other input data 33 can be from an air quality monitoring system 70 and can include weather data (temperature, wind speed, wind direction, humidity) and event data (hour of day, day of month, weekday, weekend, holiday, vacation) , and sensor data is collected by sensors and can be sent to apparatus 10 for sensor fault detection of the present disclosure.
  • sensor 21 is a target sensor (we call it “first sensor” ) to be tested whether is in fault.
  • At least one sensor 22 (we call it “second sensor” ) provides sensor data for reference. Sensor data can be received by the apparatus 10 and store in at least one memory 105. Sensor data collected by sensor 21 is called sensor data 31 and sensor data collected by each of the at least one sensor 22 is called sensor data 32.
  • the apparatus 10 for sensor fault detection presented in the present disclosure can be implemented as a network of computer processors, to execute following method 300 for sensor fault detection presented in the present disclosure.
  • the apparatus 10 can also be a single computer, as shown in FIG. 1, including at least one memory 105, which includes computer-readable medium, such as a random access memory (RAM) .
  • the apparatus 10 also includes at least one processor 104, coupled with the at least one memory 105.
  • Computer-executable instructions are stored in the at least one memory 105, and when executed by the at least one processor 104, can cause the at least one processor 104 to perform the steps described herein.
  • the at least one processor 104 may include a microprocessor, an application specific integrated circuit (ASIC) , a digital signal processor (DSP) , a central processing unit (CPU) , a graphics processing unit (GPU) , state machines, etc.
  • ASIC application specific integrated circuit
  • DSP digital signal processor
  • CPU central processing unit
  • GPU graphics processing unit
  • embodiments of computer-readable medium include, but not limited to a floppy disk, CD-ROM, magnetic disk, memory chip, ROM, RAM, an ASIC, a configured processor, all optical media, all magnetic tape or other magnetic media, or any other medium from which a computer processor can read instructions.
  • various other forms of computer-readable medium may transmit or carry instructions to a computer, including a router, private or public network, or other transmission device or channel, both wired and wireless.
  • the instructions may include code from any computer-programming language, including, for example, C, C++, C#, Visual Basic, Java, and JavaScript.
  • the at least one memory 105 shown in FIG. 1 can contain a sensor fault detection program 30, when executed by the at least one processor 104, causing the at least one processor 104 to execute the method 300 for sensor fault detection presented in the present disclosure.
  • Sensor data 31, sensor data 32 and event data 33 can also be stored in the at least one memory 105, such as weather change information. These data can be received via a communication module 106 of the apparatus 10.
  • the sensor fault detection program 40 can include:
  • a data collecting module 101 configured to collect sensor data of a first sensor 21 and of at least one second sensor 22;
  • a calculator 102 configured to calculate similarity of the collected sensor data 31 of the first sensor 21 with each of the at least one second sensor 22, regardless of influence of following two factors:
  • determining module 103 configured to determine whether the first sensor 21 is in fault based on each calculated similarity.
  • the data collecting module 101 is further configured to collect the sensor data during a first historical time period
  • the determining module 103 is further configured to determine whether the first sensor 21 is in fault during the first historical time period based on each calculated similarity.
  • the determining module 103 is further configured to compare each calculated similarity with a pre-defined threshold for the pair of the first sensor 21 and the corresponding second sensor 22 during the first historical time period, and determine whether the first sensor 21 is in fault during the first historical time period based on each comparison result.
  • the data collecting module 101 is further configured to collect sensor data of the first sensor 21 and of the at least one second sensor 22 during at least two second historical time periods; and for each second sensor, the calculator 102 is further configured to calculate, for each second historical time period, similarity of the collected sensor data of the first sensor 21 with the second sensor 22during the corresponding second historical time period; the determining module 103 is further configured to determine the pre-defined threshold for the pair of the first sensor 21 and the second sensor 22, based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
  • An air quality monitoring system in a city by a limited number of air quality monitoring sensors.
  • the air quality monitoring system located in the environmental office of the city receives the air quality data from these installed sensors somewhere outdoor in the city every hour. Now the environmental office needs to check if any sensor has fault in the last week. This an offline problem.
  • the main idea is to calculate a similarity score set between the target sensor and other sensors which are located at different places based on their historical sensor data. This similarity score is used to determine if the sensor has fault in the last week quantitatively.
  • sensors 21 and 22 transmit air quality data every hour to the historical air quality database (which can be implemented on the at least one memory 105 or be an external database) .
  • weather data to indicate an event happened and to be happened, such as weather change and weather forecast information
  • the sensor fault detection program 40 can be run every week after collecting one-week’s air quality data and weather data.
  • the program 40 can calculate a similarity score set of the target sensor 21 by comparing it with all the other sensors 22. If any score in the set fails to be greater than the threshold, the sensor is determined as fault.
  • the goal is to determine whether the targeted air quality sensor 21 is in fault or not in the last certain time period T (e.g., one week in this application scenario) , by comparing its T-length sensor data with all the other sensors 22.
  • T e.g., one week in this application scenario
  • air quality can be measured by the concentration of the monitored air pollutants of the sensors (e.g., carbon dioxide (CO2) , sulfur oxides (SOx) , nitrogen oxides (NOx) , ozone (O3) , etc.
  • CO2 carbon dioxide
  • SOx sulfur oxides
  • NOx nitrogen oxides
  • O3 ozone
  • the target sensor 21 collected one-week data with timestamp,SEQ A .
  • the sensor 21 is then compared with sensor data of another sensor B from the same time period, SEQ B and a similarity score SimScore (A, B) can be calculated, which integrates the weather data and events data:
  • the similarity function f has variables diff (A, B) , IN A , IN B , and the parameter ⁇ .
  • diff (A, B) is the difference between A and B in the same time period T, it can be:
  • sensor A (t) , sensor B (t) are the sensor’s monitoring value at t time point of sensor 21 and sensor 22 respectively;
  • p is a positive integer which is greater than 1.
  • sensor A (t) , sensor B (t) are the sensor’s monitoring value at t time point of sensor 21 and sensor 22 respectively.
  • IN A , IN B input data 33 other than sensor data , monitored at the locations of sensor 21 and sensor 22, respectively, such as the weather data (temperature, wind speed, wind direction, humidity) and event data at same time (hour of day, day of month, weekday, weekend, holiday, vacation) .
  • Integration of IN A , IN B is to evaluate the pure similarity of sensor data between sensors regardless of influence of following 2 factors:
  • parameter ⁇ can be determined by historical sensor data of sensor 21 and sensor 22, the details are explained later.
  • the certain threshold can also be determined by the minimal threshold that satisfy the condition (1) :
  • threshold argmin ⁇ Pr [ ⁇ SimScore (A′ , B) j ⁇ ⁇ threshold ] > 95% ⁇ (2)
  • FIG. 2A and FIG. 2B The solid lines represent sensor data collected by sensor A, and the dotted lines represent sensor data collected by sensor B. Both are used to measure concentration of PM1.0.
  • the horizontal axes represent time, and the vertical axes represent concentration of PM1.0.
  • the SimScore (A, B) T1 is equal to 0.99, since the changing trend of the two sensors during time period T1 is the same as the historical changing trend; when sensor A is in fault, the sensor data collected by sensor A and sensor B during time period T2 are different, the SimScore (A, B) T2 is equal to 0.3 and the changing trend is marked as fault by the historical data.
  • the probability of historical value can be considered instead of the exact value, which avoids using the average value as the threshold , weakens the influence of historical absolute value on the future trend, and makes the threshold setting more accurate and can reflect the change of time.
  • sensor data in FIG. 2A if taking average value as the threshold, difference between sensor data of sensor A and sensor B will result in determination of sensor B’s being in fault. But with method provided here, sensor B can be wrongly determined as in fault.
  • the data collecting module 101, the calculator 102 and the determining module 103 are described above as software modules of the sensor fault detection program 40. Also, they can be implemented via hardware, such as ASIC chips. They can be integrated into one chip, or separately implemented and electrically connected.
  • FIG. 1 The architecture above is merely exemplary and used to explain the exemplary method 300 shown in FIG. 3.
  • One exemplary method 300 according to the present disclosure includes following steps:
  • S301 collecting sensor data of a first sensor and of at least one second sensor
  • S302 calculating similarity of the collected sensor data of the first sensor with each of the at least one second sensor, regardless of influence of following two factors:
  • step S301 the sensor data is collected data during a first historical time period; and in step S303, the first sensor can be determined whether is in fault during the first historical time period based on each calculated similarity.
  • step S303 can include following 2 sub steps:
  • S3032 determining whether the first sensor is in fault during the first historical time period based on each comparison result.
  • the method 300 can further include following steps:
  • S304 collecting sensor data of the first sensor and of the at least one second sensor during at least two second historical time periods; and for each second sensor,
  • S305 calculating, for each second historical time period, similarity of the collected sensor data of the first sensor with the second sensor during the corresponding second historical time period;
  • S306 determining the pre-defined threshold for the pair of the first sensor and the second sensor, based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
  • FIG. 4 depicts a block diagrams of a system 400 which has similar structure with system 100.
  • a target sensor 20 in the system 400 is located somewhere in an area, and following data can be used for fault detection:
  • - second data 62 that is historical forecast data affecting sensor data, taking an example, weather data such as temperature, wind speed, wind direction and humidity can affect sensor data of concentration of air pollutant,
  • the apparatus 40 for sensor fault detection shown in FIG. 4 has similar structure with apparatus 10 shown in FIG. 1.
  • the apparatus 40 can be implemented as a network of computer processors or a single computer to execute following method 600 for sensor fault detection presented in the present disclosure.
  • the apparatus 40 can include at least one memory 405, at least one processor 404 coupled with the at least one memory 405.
  • Computer-executable instructions can be stored in the at least one memory 405, and when executed by the at least one processor 404, can cause the at least one processor 404 to perform the steps described herein.
  • the at least one memory 405 shown in FIG. 4 can contain a sensor fault detection program 60, when executed by the at least one processor 404, causing the at least one processor 404 to execute the method 600 for sensor fault detection presented in the present disclosure.
  • Data 61, 62, 63 and 64 can also be stored in the at least one memory 405. These data can be received via a communication module 406 of the apparatus 40.
  • the sensor fault detection program 60 can include:
  • a forecasting module 601 configured to forecast sensor data of a sensor based on a model, wherein the model is trained based on first data 61, second data 62 and third data 63;
  • determining module 603 configured to determine whether the sensor is in fault by comparing the fourth data 64 and the forecast sensor data during the period collecting the fourth data 64.
  • Sensors including a target sensor 20 are used for monitoring air quality. These sensors transmit air quality data every hour to the apparatus 40. Meanwhile the weather data (affecting air quality) is also acquired and updated every hour. All the air quality sensor data and weather data are needed to train an air quality forecast model. The model produces air quality forecast result for the next hours. If there is air pollution taking place, the system 600 can also generate early warnings of air pollution. The air quality sensor data collected by the target sensor 20 in the next N hours will be compared with the results forecast by the model, if the value is out of the forecast range, then sensor is reported as in fault.
  • historical forecast weather data (such as the temperature, wind speed, wind direction, humidity) , that is example of second data 62;
  • third data 63 - deviation between the historical forecast weather and real monitored/observed weather data for the past time steps (for future steps, it’s zero) , and real monitored/observed weather data is example of third data 63;
  • time point t which includes: Hour of day, day of month, weekday, weekend, holiday, vacation.
  • the real air quality values y′ t+1 , ..., y′ t+N are compared with the forecasted result y t+1 , ..., y t+N in the next N hours. If we define a Dirac function:
  • the target sensor with data y′ t is regarded as “out of range” and the sensor is reported as fault.
  • the forecasting module 601, the data collecting module 602 and the determining module 603 are described above as software modules of the sensor fault detection program 60. Also, they can be implemented via hardware, such as ASIC chips. They can be integrated into one chip, or separately implemented and electrically connected.
  • FIG. 4 The architecture above is merely exemplary and used to explain the exemplary method 600 shown in FIG. 6.
  • One exemplary method 600 according to the present disclosure includes following steps:
  • S601 forecasting sensor data of a sensor based on a model, wherein the model is trained based on first data, second data and third data;
  • S602 collecting fourth data, wherein the fourth data includes real time sensor data of the sensor;
  • S603 determining whether the sensor is in fault by comparing the fourth data and the forecast sensor data during the period collecting the fourth data.
  • the method 600 can further include following steps:
  • S604 collecting first data, which includes historical sensor data of the sensor
  • S605 collecting historical forecast data (the second data) and historical observed data (the third data) affecting sensor data of the sensor;
  • a computer-readable medium is also provided in the present disclosure, storing computer-executable instructions, which upon execution by a computer, enables the computer to execute any of the methods presented in this disclosure.
  • a computer program which is being executed by at least one processor and performs any of the methods presented in this disclosure.
  • the solutions in the present disclosure provide ways to automatically detect the fault of air quality sensors based on data-driven methods is in-time and more accurate than by eyes.
  • the models used for sensor data forecast are trained by the historical forecast data and time factors, which can handle the uncertainty of forecast data well.

Landscapes

  • Chemical & Material Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Food Science & Technology (AREA)
  • Combustion & Propulsion (AREA)
  • Immunology (AREA)
  • Medicinal Chemistry (AREA)
  • Environmental & Geological Engineering (AREA)
  • Atmospheric Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Ecology (AREA)
  • Environmental Sciences (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

A method (300, 600), apparatus (10, 40), system and computer-readable medium for sensor fault detection, to detect sensor malfunctions such as insensitiveness timely and reliably. A method (300, 600) for sensor fault detection includes: collecting sensor data of a first sensor (21) and of at least one second sensor (22)(S301); calculating similarity of the collected sensor data (31, 32), regardless of influence of event happened affecting sensor data (31, 32) and difference of sensor data (31, 32) due to location (S302); determining whether the first sensor (21) is in fault based on each calculated similarity (S303).

Description

Method and apparatus for sensor fault detection Technical Field
The present invention relates to techniques of sensor, and more particularly to a method, apparatus and computer-readable storage medium for sensor fault detection.
Background Art
A sensor/transducer is a device that responds to a physical stimulus (such as heat, light, sound, pressure, magnetism, or particular motion) and transmits a resulting impulse (as for measurement or control) , which we usually call as sensor data.
Commonly, sensors are deployed near monitored devices. Due to environmental factors, sensors may not work well, for example, being insensitive, reporting wrong data value, etc. unfortunately, such malfunctions are usually not easy to detect in comparison to no data transmission or blackout.
Summary of the Invention
Embodiments of the present disclosure include methods, apparatuses for sensor fault detection, which can detect above sensor malfunctions timely and reliably.
According to a first aspect of the present disclosure, a method, apparatus, system, computer-readable medium for sensor fault detection are presented. Sensor data are collected and similarity of sensor data from different sensors are compared regardless of influence of event happened affecting sensor data and sensor location differences. Whether a sensor is in fault is based on the calculated similarity. By comparing sensor data of different sensors, malfunctions such as insensitiveness, wrong data value reporting, etc. can be detected easily. And regardless of event influence and location differences, the detection result can be accurate and faults can be detected in time.
Optionally, sensor data can be collected during a first historical time period and the sensor can be determined whether to be in fault during the first historical time period. Each calculated similarity can be compared with a pre-defined threshold for the pair of the first sensor and the corresponding second sensor during the first historical time period, and whether the first sensor is in fault can be determined during the first historical time period based on each comparison result.
Optionally, the threshold can be determined as such: sensor data of the first sensor and of the at least one second sensor can be collected during at least two second historical time periods; and for each second sensor, similarity of the collected sensor data of the first sensor with the second sensor can be calculated for each second historical time period during the corresponding second historical time period; then the pre-defined threshold for the pair of the first sensor and the second sensor can be determined based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period. By using the method of probability instead of average, the probability of historical value can be considered instead of the exact value, which avoids using the average value as the threshold , weakens the influence of historical absolute value on the future trend, and makes the threshold setting more accurate and can reflect the change of time.
According to a second aspect of the present disclosure, a method, apparatus, system, computer-readable medium for sensor fault detection are presented. A model is trained based on historical sensor data, historical forecast data affecting sensor data and historical observed data affecting sensor data. Sensor data can be forecast based on a model, and the forecast sensor data and real time sensor data collected can be compared for determination on whether the sensor is in fault. The model used for sensor data forecast is trained based on the historical forecast data and time factors, which can handle the uncertainty of forecast data well.
Brief Description of the Drawings
The above mentioned attributes and other features and advantages of the present technique and the manner of attaining them will become more apparent and the present technique itself will be better understood by reference to the following description of embodiments of the present technique taken in conjunction with the accompanying drawings, wherein:
FIG. 1 depicts a block diagram of a system for sensor fault detection.
FIG. 2A and FIG. 2B depict similarity of sensor under normal status and fault status.
FIG. 3 depicts a flow diagram of a method for sensor fault detection in accordance with one embodiment of the present disclosure.
FIG. 4 depicts a block diagram of a system for sensor fault detection.
FIG. 5 depicts structure of a model for forecasting sensor data.
FIG. 6 depicts a flow diagram of a method for sensor fault detection in accordance with one embodiment of the present disclosure.
Reference Numbers:
100, a system for sensor fault detection
10, an apparatus for sensor fault detection
70, an air quality monitoring system
21, a first sensor
22, a second sensor
30, a sensor fault detection program
31, sensor data of a first sensor 21
32, sensor data of a second sensor 22
33, input data other than  sensor data  31 and 32
101, a data collecting module
102, a calculator
103, a determining module
104, at least one processor
105, at least one memory
106, a communication module
S301, collecting sensor data
S302, calculating similarity
S303, fault determination
S3031, comparing similarity
S3032, fault determination
S304, collecting sensor data
S305, calculating similarity
S306, threshold determination
400, a system for sensor fault detection
40, an apparatus for sensor fault detection
20, a sensor
60, a sensor fault detection program
61, first data
62, second data
63, third data
64, fourth data
401, a forecasting module
402, a data collecting module
403, a determining module
404, at least one processor
405, at least one memory
406, a communication module
S601, forecasting sensor data
S602, collecting real time sensor data
S603, fault determination
S604, collecting historical sensor data
S605, collecting historical forecast data and observed data affecting sensor data
S606, train a model
Detailed Description of Example Embodiments
Hereinafter, above-mentioned and other features of the present technique are described in detail. Various embodiments are described with reference to the drawing, where like reference numerals are used to refer to like elements throughout. In the following description, for purpose of explanation, numerous specific details are set forth in order to provide a thorough understanding of one or more embodiments. It may be noted that the illustrated embodiments are intended to explain, and not to limit the invention. It may be evident that such embodiments may be practiced without these specific details.
When introducing elements of various embodiments of the present disclosure, the articles “a” , “an” , “the” and “said” are intended to mean that there are one or more of the elements. The terms “comprising” , “including” and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements.
Now the present disclosure will be described hereinafter in details by referring to FIG. 1 to FIG. 6.
FIG. 1 depicts a block diagrams of a system 100 in accordance with one embodiment of the present disclosure. As shown in FIG. 1, there are sensors located on different places in an area, such as a city, a district in a city, a community or a building.
For example, these sensors are used for monitoring air quality of a city, there  are  sensor data  31, 32 and other input data 33. The other input data 33 can be from an air quality monitoring system 70 and can include weather data (temperature, wind speed, wind direction, humidity) and event data (hour of day, day of month, weekday, weekend, holiday, vacation) , and sensor data is collected by sensors and can be sent to apparatus 10 for sensor fault detection of the present disclosure.
Here, sensor 21 is a target sensor (we call it “first sensor” ) to be tested whether is in fault. At least one sensor 22 (we call it “second sensor” ) provides sensor data for reference. Sensor data can be received by the apparatus 10 and store in at least one memory 105. Sensor data collected by sensor 21 is called sensor data 31 and sensor data collected by each of the at least one sensor 22 is called sensor data 32.
The apparatus 10 for sensor fault detection presented in the present disclosure can be implemented as a network of computer processors, to execute following method 300 for sensor fault detection presented in the present disclosure. the apparatus 10 can also be a single computer, as shown in FIG. 1, including at least one memory 105, which includes computer-readable medium, such as a random access memory (RAM) . The apparatus 10 also includes at least one processor 104, coupled with the at least one memory 105. Computer-executable instructions are stored in the at least one memory 105, and when executed by the at least one processor 104, can cause the at least one processor 104 to perform the steps described herein. The at least one processor 104 may include a microprocessor, an application specific integrated circuit (ASIC) , a digital signal processor (DSP) , a central processing unit (CPU) , a graphics processing unit (GPU) , state machines, etc. embodiments of computer-readable medium include, but not limited to a floppy disk, CD-ROM, magnetic disk, memory chip, ROM, RAM, an ASIC, a configured processor, all optical media, all magnetic tape or other magnetic media, or any other medium from which a computer processor can read instructions. Also, various other forms of computer-readable medium may transmit or carry instructions to a computer, including a router, private or public network, or other transmission device or channel, both wired and wireless. The instructions may include code from any computer-programming language, including, for example, C, C++, C#, Visual Basic, Java, and JavaScript.
The at least one memory 105 shown in FIG. 1 can contain a sensor fault detection program 30, when executed by the at least one processor 104, causing the at least one processor 104 to execute the method 300 for sensor fault detection presented in the present disclosure. Sensor data 31, sensor data 32 and event data 33 can also be stored in the at least one memory 105, such as weather change  information. These data can be received via a communication module 106 of the apparatus 10.
The sensor fault detection program 40 can include:
- a data collecting module 101, configured to collect sensor data of a first sensor 21 and of at least one second sensor 22;
- a calculator 102, configured to calculate similarity of the collected sensor data 31 of the first sensor 21 with each of the at least one second sensor 22, regardless of influence of following two factors:
- event happened affecting sensor data of the first sensor 21 and of the at least one second sensor 22;
- difference of sensor data 31 of the first sensor 21 due to location differences from the at least one second sensor 22;
- a determining module 103, configured to determine whether the first sensor 21 is in fault based on each calculated similarity.
Optionally, the data collecting module 101 is further configured to collect the sensor data during a first historical time period, and the determining module 103 is further configured to determine whether the first sensor 21 is in fault during the first historical time period based on each calculated similarity.
Optionally, the determining module 103 is further configured to compare each calculated similarity with a pre-defined threshold for the pair of the first sensor 21 and the corresponding second sensor 22 during the first historical time period, and determine whether the first sensor 21 is in fault during the first historical time period based on each comparison result.
Optionally, the data collecting module 101 is further configured to collect sensor data of the first sensor 21 and of the at least one second sensor 22 during at least two second historical time periods; and for each second sensor, the calculator 102 is further configured to calculate, for each second historical time period, similarity of the collected sensor data of the first sensor 21 with the second sensor 22during the corresponding second historical time period; the determining module 103 is further configured to determine the pre-defined threshold for the pair of the first sensor 21 and the second sensor 22, based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
Following is an example:
An air quality monitoring system in a city by a limited number of air quality monitoring sensors. The air quality monitoring system located in the environmental  office of the city receives the air quality data from these installed sensors somewhere outdoor in the city every hour. Now the environmental office needs to check if any sensor has fault in the last week. This an offline problem.
The main idea is to calculate a similarity score set between the target sensor and other sensors which are located at different places based on their historical sensor data. This similarity score is used to determine if the sensor has fault in the last week quantitatively.
Referring to FIG. 1,  sensors  21 and 22 transmit air quality data every hour to the historical air quality database (which can be implemented on the at least one memory 105 or be an external database) . Meanwhile weather data (to indicate an event happened and to be happened, such as weather change and weather forecast information) can also be acquired and updated every hour. The sensor fault detection program 40 can be run every week after collecting one-week’s air quality data and weather data. Furthermore, the program 40 can calculate a similarity score set of the target sensor 21 by comparing it with all the other sensors 22. If any score in the set fails to be greater than the threshold, the sensor is determined as fault.
The goal is to determine whether the targeted air quality sensor 21 is in fault or not in the last certain time period T (e.g., one week in this application scenario) , by comparing its T-length sensor data with all the other sensors 22.
This is to explain how to compute the similarity score and determine the threshold. Quantitatively, air quality can be measured by the concentration of the monitored air pollutants of the sensors (e.g., carbon dioxide (CO2) , sulfur oxides (SOx) , nitrogen oxides (NOx) , ozone (O3) , etc.
In particular, the target sensor 21 collected one-week data with timestamp,SEQ A. The sensor 21 is then compared with sensor data of another sensor B from the same time period, SEQ B and a similarity score SimScore (A, B) can be calculated, which integrates the weather data and events data:
SimScore (A, B) =f ( [diff (A, B) ] , IN A, IN B|θ)
The similarity function f has variables diff (A, B) , IN A, IN B, and the parameter θ.
diff (A, B) is the difference between A and B in the same time period T, it can be:
1) |Area (SEQ A) - Area (SEQ B) |, where Area (SEQ A) , Area (SEQ B) are the areas under the curve of one-week corresponding air pollutant data from sensor A and sensor B.
2) 
Figure PCTCN2020073539-appb-000001
where sensor A (t) , sensor B (t) are the sensor’s monitoring value at t time point of sensor 21 and sensor 22 respectively; p is a positive integer which is greater than 1.
3) 
Figure PCTCN2020073539-appb-000002
where sensor A (t) , sensor B (t) are the sensor’s monitoring value at t time point of sensor 21 and sensor 22 respectively.
IN A, IN B: input data 33 other than sensor data , monitored at the locations of sensor 21 and sensor 22, respectively, such as the weather data (temperature, wind speed, wind direction, humidity) and event data at same time (hour of day, day of month, weekday, weekend, holiday, vacation) . Integration of IN A, IN B is to evaluate the pure similarity of sensor data between sensors regardless of influence of following 2 factors:
- event happened affecting sensor data of the first sensor and of the at least one second sensor;
- difference of sensor data of the first sensor due to location differences from the at least one second sensor;
parameter θ can be determined by historical sensor data of sensor 21 and sensor 22, the details are explained later.
By computing all the similarity scores from N-week historical data of sensor A and sensor B: {SimScore (A, B)  kk=1, …N, the N scores can be ranked based on the values. The best similarity function
Figure PCTCN2020073539-appb-000003
based on the determined parameter
Figure PCTCN2020073539-appb-000004
should satisfy:
Figure PCTCN2020073539-appb-000005
where the sensor A′ means that sensor A has fault at certain week j in history. SimScore (A′ , B)  j represents the similarity score between the fault sensor A′ and normal sensor B at certain week j. In this way, the certain threshold can also be determined by the minimal threshold that satisfy the condition (1) :
threshold : = argmin {Pr [ {SimScore (A′ , B)  j} < threshold ] > 95%}   (2)
Here, 95%is just an example, it can be set according to actual requirements. Now referring to FIG. 2A and FIG. 2B. The solid lines represent sensor data collected by sensor A, and the dotted lines represent sensor data collected by sensor B. Both are used to measure concentration of PM1.0. The horizontal axes represent time, and the vertical axes represent concentration of PM1.0. When both sensors are in normal status (referring to FIG. 2A) , the sensor data collected by sensor A and sensor B during time period T1 are similar, the SimScore (A, B)  T1 is equal to 0.99,  since the changing trend of the two sensors during time period T1 is the same as the historical changing trend; when sensor A is in fault, the sensor data collected by sensor A and sensor B during time period T2 are different, the SimScore (A, B)  T2 is equal to 0.3 and the changing trend is marked as fault by the historical data.
Here, By using the method of probability instead of average, the probability of historical value can be considered instead of the exact value, which avoids using the average value as the threshold , weakens the influence of historical absolute value on the future trend, and makes the threshold setting more accurate and can reflect the change of time. Taking sensor data in FIG. 2A as example, if taking average value as the threshold, difference between sensor data of sensor A and sensor B will result in determination of sensor B’s being in fault. But with method provided here, sensor B can be wrongly determined as in fault.
Although the data collecting module 101, the calculator 102 and the determining module 103 are described above as software modules of the sensor fault detection program 40. Also, they can be implemented via hardware, such as ASIC chips. They can be integrated into one chip, or separately implemented and electrically connected.
It should be mentioned that the present disclosure may include apparatuses having different architecture than shown in FIG. 1. The architecture above is merely exemplary and used to explain the exemplary method 300 shown in FIG. 3.
Various methods in accordance with the present disclosure may be carried out. One exemplary method 300 according to the present disclosure includes following steps:
S301: collecting sensor data of a first sensor and of at least one second sensor;
S302: calculating similarity of the collected sensor data of the first sensor with each of the at least one second sensor, regardless of influence of following two factors:
- event happened affecting sensor data of the first sensor and of the at least one second sensor;
- difference of sensor data of the first sensor due to location differences from the at least one second sensor;
S303: determining whether the first sensor is in fault based on each calculated similarity.
Optionally, in step S301, the sensor data is collected data during a first historical time period; and in step S303, the first sensor can be determined whether is in fault during the first historical time period based on each calculated similarity.
Optionally, the step S303 can include following 2 sub steps:
S3031: comparing each calculated similarity with a pre-defined threshold for the pair of the first sensor and the corresponding second sensor during the first historical time period;
S3032: determining whether the first sensor is in fault during the first historical time period based on each comparison result.
Optionally, the method 300 can further include following steps:
S304: collecting sensor data of the first sensor and of the at least one second sensor during at least two second historical time periods; and for each second sensor,
S305: calculating, for each second historical time period, similarity of the collected sensor data of the first sensor with the second sensor during the corresponding second historical time period;
S306: determining the pre-defined threshold for the pair of the first sensor and the second sensor, based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
With the above-mentioned system 100, apparatus 10 and method 300, similarity between sensors can be calculated, which serves as basis for sensor fault detection. Now referring to FIG. 4 to 6, the other solution for sensor fault detection is provided, in which sensor data is forecast and compared with collected sensor data, for sensor fault determination.
FIG. 4 depicts a block diagrams of a system 400 which has similar structure with system 100. A target sensor 20 in the system 400 is located somewhere in an area, and following data can be used for fault detection:
first data 61, that is historical sensor data,
second data 62, that is historical forecast data affecting sensor data, taking an example, weather data such as temperature, wind speed, wind direction and humidity can affect sensor data of concentration of air pollutant,
third data 63, that is historical observed data affecting sensor data,
fourth data 64, that is real time sensor data.
The apparatus 40 for sensor fault detection shown in FIG. 4 has similar structure with apparatus 10 shown in FIG. 1. The apparatus 40 can be implemented as a network of computer processors or a single computer to execute following method 600 for sensor fault detection presented in the present disclosure. When the  apparatus 40 is implemented as a single computer, it can have similar structure and implementation with apparatus 10 shown in FIG. 1. The apparatus 40 can include at least one memory 405, at least one processor 404 coupled with the at least one memory 405. Computer-executable instructions can be stored in the at least one memory 405, and when executed by the at least one processor 404, can cause the at least one processor 404 to perform the steps described herein.
The at least one memory 405 shown in FIG. 4 can contain a sensor fault detection program 60, when executed by the at least one processor 404, causing the at least one processor 404 to execute the method 600 for sensor fault detection presented in the present disclosure.  Data  61, 62, 63 and 64 can also be stored in the at least one memory 405. These data can be received via a communication module 406 of the apparatus 40.
The sensor fault detection program 60 can include:
- a forecasting module 601, configured to forecast sensor data of a sensor based on a model, wherein the model is trained based on first data 61, second data 62 and third data 63;
- a data collecting module 602, configured to collect fourth data 64;
- a determining module 603, configured to determine whether the sensor is in fault by comparing the fourth data 64 and the forecast sensor data during the period collecting the fourth data 64.
Following is an example:
Sensors including a target sensor 20 are used for monitoring air quality. These sensors transmit air quality data every hour to the apparatus 40. Meanwhile the weather data (affecting air quality) is also acquired and updated every hour. All the air quality sensor data and weather data are needed to train an air quality forecast model. The model produces air quality forecast result for the next hours. If there is air pollution taking place, the system 600 can also generate early warnings of air pollution. The air quality sensor data collected by the target sensor 20 in the next N hours will be compared with the results forecast by the model, if the value is out of the forecast range, then sensor is reported as in fault.
Next, how to compute the forecast air quality results for the next N hours is introduced. Regardless of types of air pollutants here, we are using yt (air pollutant) to represent the concentration of the air pollutant at time point t; using IN t to represent the model input, including:
- air quality sensor data at the last time point, that is example of first data 61;
- historical forecast weather data (such as the temperature, wind speed, wind direction, humidity) , that is example of second data 62;
- deviation between the historical forecast weather and real monitored/observed weather data for the past time steps (for future steps, it’s zero) , and real monitored/observed weather data is example of third data 63;
- events data at time point t which includes: Hour of day, day of month, weekday, weekend, holiday, vacation.
By using the model structure shown in FIG. 5, the AQF model can be explained by the following formula:
S t+1=tanh(A·S t+B·IN t)    (3)
y t+1=C·S t+1                 (4)
This is an RNN (Recurrent Neural Network) model and using BPTT (back propagation through time) (Werbos, Paul J. (1988) . "Generalization of backpropagation with application to a recurrent gas market model" . Neural Networks. 1 (4) : 339–356. doi: 10.1016/0893-6080 (88) 90007-x) to train the matrix A, B and C in equation (3) and equation (4) . Once A, B and C are computed, the air quality of next hour y t+1 can be computed.
Next, in order to determine if the target sensor 20 is in fault or not, the real air quality values y′ t+1, …, y′ t+N are compared with the forecasted result y t+1, …, y t+N in the next N hours. If we define a Dirac function:
Figure PCTCN2020073539-appb-000006
If
δ (t) δ (t+1) …δ (t+N) = 1,
then the target sensor with data y′ t is regarded as “out of range” and the sensor is reported as fault.
Similar to apparatus 10, although the forecasting module 601, the data collecting module 602 and the determining module 603 are described above as software modules of the sensor fault detection program 60. Also, they can be implemented via hardware, such as ASIC chips. They can be integrated into one chip, or separately implemented and electrically connected.
It should be mentioned that the present disclosure may include apparatuses having different architecture than shown in FIG. 4. The architecture above is merely exemplary and used to explain the exemplary method 600 shown in FIG. 6.
Various methods in accordance with the present disclosure may be carried out. One exemplary method 600 according to the present disclosure includes following steps:
S601: forecasting sensor data of a sensor based on a model, wherein the model is trained based on first data, second data and third data;
S602: collecting fourth data, wherein the fourth data includes real time sensor data of the sensor;
S603: determining whether the sensor is in fault by comparing the fourth data and the forecast sensor data during the period collecting the fourth data.
Optionally, before the step S601, the method 600 can further include following steps:
S604: collecting first data, which includes historical sensor data of the sensor;
S605: collecting historical forecast data (the second data) and historical observed data (the third data) affecting sensor data of the sensor;
S606: training the model based on the first data, the second data and the third data.
A computer-readable medium is also provided in the present disclosure, storing computer-executable instructions, which upon execution by a computer, enables the computer to execute any of the methods presented in this disclosure.
A computer program, which is being executed by at least one processor and performs any of the methods presented in this disclosure.
The solutions in the present disclosure provide ways to automatically detect the fault of air quality sensors based on data-driven methods is in-time and more accurate than by eyes. The models used for sensor data forecast are trained by the historical forecast data and time factors, which can handle the uncertainty of forecast data well.
While the present technique has been described in detail with reference to certain embodiments, it should be appreciated that the present technique is not limited to those precise embodiments. Rather, in view of the present disclosure  which describes exemplary modes for practicing the invention, many modifications and variations would present themselves, to those skilled in the art without departing from the scope and spirit of this invention. The scope of the invention is, therefore, indicated by the following claims rather than by the foregoing description. All changes, modifications, and variations coming within the meaning and range of equivalency of the claims are to be considered within their scope.

Claims (13)

  1. A method (300) for sensor fault detection, comprising:
    -collecting (S301) sensor data of a first sensor and of at least one second sensor; 
    -calculating (S302) similarity of the collected sensor data of the first sensor with each of the at least one second sensor, regardless of influence of following two factors:
    -event happened affecting sensor data of the first sensor and of the at least one second sensor;
    -difference of sensor data of the first sensor due to location differences from the at least one second sensor;
    -determining (S303) whether the first sensor is in fault based on each calculated similarity.
  2. the method (300) according to claim 1, wherein,
    -collecting (S301) sensor data of a first sensor and of at least one second sensor, comprises: collecting the sensor data during a first historical time period;
    -determining (S303) whether the first sensor is in fault, comprises: determining whether the first sensor is in fault during the first historical time period based on each calculated similarity.
  3. the method (300) according to claim 2, wherein determining (S303) whether the first sensor is in fault during the first historical time period based on each calculated similarity, comprises:
    -comparing (S3031) each calculated similarity with a pre-defined threshold for the pair of the first sensor and the corresponding second sensor during the first historical time period;
    -determining (S3032) whether the first sensor is in fault during the first historical time period based on each comparison result.
  4. the method (300) according to claim 3, further comprising:
    -collecting (S304) sensor data of the first sensor and of the at least one second sensor during at least two second historical time periods;
    -for each second sensor,
    -calculating (S305) , for each second historical time period, similarity of the collected sensor data of the first sensor with the second sensor during the corresponding second historical time period;
    -determining (S306) the pre-defined threshold for the pair of the first sensor and the second sensor, based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
  5. A method (600) for sensor fault detection, comprising:
    -forecasting (S601) sensor data of a sensor based on a model, wherein the model is trained based on first data, second data and third data, the first data comprises historical sensor data of the sensor, the second data comprises historical forecast data affecting sensor data of the sensor, the third data comprises historical observed data affecting sensor data of the sensor;
    -collecting (S602) fourth data, wherein the fourth data comprises real time sensor data of the sensor;
    -determining (S603) whether the sensor is in fault by comparing the fourth data and the forecast sensor data during the period collecting the fourth data.
  6. An apparatus (10) for sensor fault detection, comprising:
    -a data collecting module (101) , configured to collect sensor data of a first sensor and of at least one second sensor;
    -a calculator (102) , configured to calculate similarity of the collected sensor data of the first sensor with each of the at least one second sensor, regardless of influence of following two factors:
    -event happened affecting sensor data of the first sensor and of the at least one second sensor;
    -difference of sensor data of the first sensor due to location differences from the at least one second sensor;
    -a determining module (103) , configured to determine whether the first sensor is in fault based on each calculated similarity.
  7. the apparatus (10) according to claim 6, wherein,
    -the data collecting module (101) is further configured to collect the sensor data during a first historical time period;
    -the determining module (103) is further configured to determine whether the first sensor is in fault during the first historical time period based on each calculated similarity.
  8. the apparatus (10) according to claim 7, wherein the determining module (103) is further configured to:
    -compare each calculated similarity with a pre-defined threshold for the pair of the first sensor and the corresponding second sensor during the first historical time period;
    -determine whether the first sensor is in fault during the first historical time period based on each comparison result.
  9. the apparatus (10) according to claim 8, wherein,
    -the data collecting module (101) is further configured to collect sensor data of the first sensor and of the at least one second sensor during at least two second historical time periods;
    -for each second sensor,
    -the calculator (102) is further configured to: calculate, for each second historical time period, similarity of the collected sensor data of the first sensor with the second sensor during the corresponding second historical time period;
    -the determining module (103) is further configured to determine the pre-defined threshold for the pair of the first sensor and the second sensor, based on each calculated similarity during respective second historical time period and real status of the first sensor during each second historical time period.
  10. An apparatus (40) for sensor fault detection, comprising:
    -a forecasting module (401) , configured to forecast sensor data of a sensor based on a model, wherein the model is trained based on first data, second data and third data, the first data comprises historical sensor data of the sensor, the second data comprises historical forecast data affecting sensor data of the sensor, the third data comprises historical observed data affecting sensor data of the sensor;
    -a data collecting module (402) , configured to collect fourth data, wherein the fourth data comprises real time sensor data of the sensor;
    -a determining module (403) , configured to determine whether the sensor is in fault by comparing the fourth data and the forecast sensor data during the period collecting the fourth data.
  11. An apparatus (10) for sensor fault detection, comprising:
    -at least one processor (104) ;
    -at least one memory (105) , coupled to the at least one processor (104) ,  configured to execute method according to any of claims 1~5.
  12. An apparatus (40) for sensor fault detection, comprising:
    -at least one processor (404) ;
    -at least one memory (405) , coupled to the at least one processor (404) , configured to execute method according claim 5.
  13. A computer-readable medium for sensor fault detection, storing computer-executable instructions, wherein the computer-executable instructions when executed cause at least one processor to execute method according to any of claims 1~5.
PCT/CN2020/073539 2020-01-21 2020-01-21 Method and apparatus for sensor fault detection WO2021146927A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/073539 WO2021146927A1 (en) 2020-01-21 2020-01-21 Method and apparatus for sensor fault detection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/073539 WO2021146927A1 (en) 2020-01-21 2020-01-21 Method and apparatus for sensor fault detection

Publications (1)

Publication Number Publication Date
WO2021146927A1 true WO2021146927A1 (en) 2021-07-29

Family

ID=76992687

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/073539 WO2021146927A1 (en) 2020-01-21 2020-01-21 Method and apparatus for sensor fault detection

Country Status (1)

Country Link
WO (1) WO2021146927A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119290081A (en) * 2024-12-10 2025-01-10 天津市热电有限公司 A multi-sensor based equipment operation status monitoring system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020042694A1 (en) * 2000-03-23 2002-04-11 Henry Manus P. Process monitoring and control using self-validating sensors
CN1768311A (en) * 2003-03-31 2006-05-03 英国电讯有限公司 Data Analysis Systems and Methods
CN101266273A (en) * 2008-05-12 2008-09-17 徐立军 Multi- sensor system fault self-diagnosis method
CN102538859A (en) * 2011-05-19 2012-07-04 广东迅通科技股份有限公司 Method for monitoring and processing various sensors
CN106961349A (en) * 2017-02-20 2017-07-18 江苏大学 A kind of sensor fault identifying system and method based on data fusion
WO2019011877A1 (en) * 2017-07-10 2019-01-17 Siemens Aktiengesellschaft Monitoring device and method for monitoring a system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020042694A1 (en) * 2000-03-23 2002-04-11 Henry Manus P. Process monitoring and control using self-validating sensors
CN1768311A (en) * 2003-03-31 2006-05-03 英国电讯有限公司 Data Analysis Systems and Methods
CN101266273A (en) * 2008-05-12 2008-09-17 徐立军 Multi- sensor system fault self-diagnosis method
CN102538859A (en) * 2011-05-19 2012-07-04 广东迅通科技股份有限公司 Method for monitoring and processing various sensors
CN106961349A (en) * 2017-02-20 2017-07-18 江苏大学 A kind of sensor fault identifying system and method based on data fusion
WO2019011877A1 (en) * 2017-07-10 2019-01-17 Siemens Aktiengesellschaft Monitoring device and method for monitoring a system

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119290081A (en) * 2024-12-10 2025-01-10 天津市热电有限公司 A multi-sensor based equipment operation status monitoring system

Similar Documents

Publication Publication Date Title
US7415328B2 (en) Hybrid model based fault detection and isolation system
JP5278310B2 (en) Diagnostic system
US11729589B2 (en) Systems and methods for verifying reliability of sensor data received from mobile devices
KR20070022996A (en) Vehicle Management System and Method of Telematics System
US9299051B2 (en) Methods and systems for evaluating the performance of building processes
KR101328889B1 (en) Structural health monitoring system based on measured displacement
KR20240011360A (en) Method and device for estimating fine dust concentration and inferencing source using local public data
CN114408694A (en) Elevator fault prediction system and prediction method thereof
WO2021146927A1 (en) Method and apparatus for sensor fault detection
Li et al. Meteorological radar fault diagnosis based on deep learning
CN119471552A (en) A method and device for self-detecting fault of electric energy meter
CN116681307A (en) Method and system for retrospective display of river disorder supervision based on multi-terminal fusion feedback
CN113965618B (en) Abnormal track detection method based on fuzzy theory
CN115639826A (en) Robot driving track deviation rectifying method and system
US20240281717A1 (en) Method and Apparatus for Training a Model
Pawling et al. WIPER: An emergency response system
CN110162902B (en) A method for reliability assessment of power distribution system based on cloud computing
JPH0822592A (en) Sudden event detection method and sudden event detection device
CN119150191B (en) Pressure switch fatigue test system and method based on data analysis
CN114999119B (en) Restricted space gas alarm system and method based on internet of things transmission function
CN117499621B (en) Detection method, device, equipment and medium of video acquisition equipment
CN116846892B (en) Intelligent energy consumption supervision system and method applying edge computing
JP7634793B1 (en) People flow measurement system, people flow measurement device, people flow measurement method, and people flow measurement program
CN114578833B (en) Forest fire prevention robot patrol method and device, electronic equipment and storage medium
CN119046545A (en) Portal service navigation method and system based on cloud platform

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20916195

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20916195

Country of ref document: EP

Kind code of ref document: A1