RU2787310C1 - Onboard analytical complex for vehicles - Google Patents

Abstract

FIELD: automotive industry,

SUBSTANCE: invention relates to a device for diagnosing and monitoring the technical condition of vehicles. The substacne of the onboard analytical complex is as follows. As a sub-block of data on the allowable, nominal and limiting values of parameters, a sub-block of triggers is used with the ability to evaluate a set of parameters and generate alerts. The driver's monitor is used as an information and control panel. To obtain the first and second types of data from the vehicle, telemetry sensors are used with the corresponding nodes for pre-processing the incoming data. In this case, the analyzed information enters the predictive analytics block, equipped with the N-number of predictive analytics models.

EFFECT: complex monitoring of the state of the vehicle in real time, prevention of critical failures of the vehicle, forecasting changes in the state of the vehicle.

4 cl, 1 dwg

Description

The invention relates to a device for remote diagnostics and real-time monitoring of the technical condition of a fleet of vehicles.

A diagnostic system and a diagnostic method are known (patent WO 2021079438 A1, G01M 17/08, G01M 99/00, publ. 29.04.2021). The aim of the present invention is to provide a diagnostic system with which it is possible to increase the possibility of realizing a highly accurate analysis and/or an efficient wear analysis of a device. The diagnostic system according to the present invention comprises: a data acquisition unit that obtains a wear value indicating the wear rate of a device installed in a vehicle, the wear information corresponding to each of a plurality of calculated days/times. Based on the measurement data, in which the state of the device is measured at each of the days/times, and traffic information, including route information indicating the route traveled by the vehicle; and a wear analysis unit that obtains first information indicative of results of the vehicle running state using the route information, and analyzes the wear of the device using the first information and wear amount over a plurality of days/times.

The disadvantage is only the implementation of monitoring the amount of wear, i.e. a decrease in the performance of the product. In addition, the system does not provide prediction of the state of the vehicle.

An anomaly detection device, an anomaly detection method and a computer program are known (patent WO 2019016996 A1, B61L 15/00, B61L 27/00, publ. 24.01.2019). Embodiments of the present invention provide highly accurate anomaly detection. According to one embodiment, the anomaly detecting device includes: a condition generator, a threshold setting device, and an anomaly detector. The generator generates a plurality of conditions for classifying a difference between a predicted vehicle state value and a measured vehicle state value based on the vehicle motion information, wherein the predicted value is based on a control command value and a prediction model. The threshold setting device sets a plurality of thresholds for conditions. The anomaly detector performs anomaly detection on the vehicle based on the prediction model, thresholds, and conditions.

This device implements only one of the predictive analytics strategies and does not allow determining the remaining resource of the node, the probability of failure in the specified time interval, and the probability of failure by operating time or by mileage.

A method is known for predicting the expected service life of a component part of an observed vehicle and a processing unit, characterized in that data on the condition of the components of selected vehicles are transmitted to the processing unit (patent RU 2720651, G01M 99/00, G01D 21/00, publ. 12.05. 2020). The operating parameter for each of the components of the selected vehicles is fed into the processing unit, while the operating parameter affects the status data of the corresponding selected component. The processing unit determines the relationship between the operating parameter and the status data for each of the selected components. The one relationship that best fits the component of the observed vehicle is selected by an algorithm. Next, the operating parameter of the component of the observed vehicle is transmitted to the processing unit. In addition, the processing unit predicts the expected service life of the observed vehicle component using the selected relationship and the operating parameter of the observed vehicle component.

The disadvantage is limited functionality - the system does not allow to determine the remaining resource of the node and the probability of failure by the time of operation or by mileage, anomaly detection is also not implemented.

Known system for predicting faults for vehicles containing sensors associated with individual components of multiple vehicles (patent GB 2378248 A, G05B 23/02, G01D 3/08, G01M 17/00, publ. 05.02.2003). The component data is collected and passed to the analysis tools, where it is analyzed and two models are created. The data is compared to these models to establish the functional state of the components and predict future failure. The "normal performance" model represents the normal operation of the vehicle and each component is assigned a range of acceptable measurement values. Data outside this range indicates that an error has occurred. The knowledge-based model provides information about the time between the receipt of a certain measurement value from a certain component and the failure of that component. All new received data is used to update the models. The data transmission may be wireless and may be carried out directly from the sensors to the analysis means or pass through a data receiver associated with each vehicle. An embodiment shows use on trains where a number of railway vehicle components (eg, car doors, brakes, heating, etc.) are controlled. A method for predicting a fault is also disclosed.

The disadvantage is limited functionality. The system does not allow to determine the remaining resource of the node and carry out full monitoring of the state of vehicles.

A known method and device for analyzing vehicle malfunctions (patent EP 3865963 A1, G05B 23/02, G07C 5/08, publ. 18.08.2021). The method includes: obtaining information about the fault to be diagnosed; obtaining from the database predetermined characteristics related to the fault to be diagnosed, which is indicated in the fault information to be diagnosed; processing, with the fault analysis model, the fault information to be diagnosed and predetermining the extracted characteristics, the fault analysis model being the preferred model determined by training the merger model using the fault history data and fault-related history characteristics based on a classification algorithm ; and generating a fault diagnosis result for the fault being diagnosed.

The disadvantage is limited functionality. The system will only allow monitoring the current state of the vehicle, there is no possibility of predicting the state.

Known on-board diagnostics in real time based on the output parameters of the commercial fleet maintenance alert system (patent US 2019066407 A1, G06Q 50/30, G07C 5/00, G07C 5/08, publ. 28.02.2019). The Commercial Fleet Maintenance Alert System is also capable of performing machine-level pattern analysis to correlate a previously reported need for a specific vehicle maintenance with a subsequent breakdown of a particular vehicle in the commercial fleet to predict the likelihood of similar future breakdowns of other vehicles in the commercial fleet. . In addition, the commercial fleet maintenance alert system can generate maintenance status reports and estimate maintenance costs for machine-identified vehicle maintenance needs, which are then compared with actual maintenance costs to improve the accuracy of future cost estimates.

The disadvantage is the limited functionality, namely the lack of determination of the residual resource of the node, the probability of failure by time of operation or by mileage and anomaly detection.

The closest technical solution, chosen as a prototype, is a device for on-board monitoring of the technical condition of the vehicle (patent RU 2631750, G07C 5/08, publ. 26.09.2017). The device contains a visual information retrieval unit, an acoustic information retrieval unit, a speed sensor, a sensor unit for monitoring engine technical condition parameters, a signal pre-processing unit, an analog-to-digital converter, a timer, a memory unit, a state prediction sub-unit, a data sub-unit on permissible nominal and limit values. parameters, a sub-block of values of the parameters of the technical condition of the vehicle, a block for pre-processing digital information, a block for transmitting information over a radio channel, a buzzer and a block for blocking the vehicle. The device is additionally equipped with a video / audio stream processing unit, a memory unit with the ability to write to an external flash drive, a pre-failure technical condition analysis sub-unit, a sub-unit for evaluating the actual operability of the vehicle, a sub-unit for a database of faults and recommendations.

The disadvantage of this system is the limited amount of information for diagnosing the state of the vehicle - the system analyzes only the speed data and parameters of the technical condition of the engine. The system does not allow to determine the probability of failure in the specified time interval, the probability of failure by the time of operation or by mileage, and also to detect anomalies. The system does not have the possibility of long-term storage of large amounts of data on the parameters of the functioning of the vehicle components and the results of the system. The system does not have a dispatch control center that allows real-time assessment of the state of the fleet of vehicles remotely.

The objective of the invention is the implementation of a comprehensive monitoring of the state of vehicles in real time, the prevention of critical transport failures and prompt response to their occurrence, as well as predicting a change in technical condition that can affect the performance of vehicle systems and traffic safety. All this helps to minimize repair and maintenance costs, reduce equipment downtime, and reduce the time spent on maintenance.

The technical result is achieved by the fact that the system is additionally equipped with

– a sensor for receiving the first type of data from the vehicle (telemetry data);

– a sensor for receiving the second type of data from the vehicle (telemetry data);

– an on-board analytical unit, including a sub-unit for preliminary processing of sensor data 1, a sub-unit for preliminary processing of sensor data 2, a sub-unit for temporary storage and transmission of data, a sub-unit for assessing the actual performance of the vehicle;

- a sub-block for assessing the actual performance of the vehicle;

- control room;

- block of data storage and structuring;

- telemetry database;

- database of historical data on breakdowns;

- database of faults and recommendations;

- vehicle condition monitoring unit;

- subblock of triggers;

- a predictive analytics lock based on machine learning;

- N-number of predictive analytics models;

- a sub-block of prediction and forecasting based on the model;

- sub-block of interpretation of results and recommendations for decision-making.

Sensors for receiving the first and second types of data from the vehicle (telemetry data) allow obtaining information about a large number of parameters of the nodes, devices and systems of the vehicle. This allows for complex monitoring, diagnostics and forecasting.

The on-board analytical unit, operating on the basis of the sub-unit for evaluating the actual performance of the vehicle, allows you to monitor the state of the vehicle in real time and promptly displays the information received to the driver or dispatcher for taking action.

The block for storing and structuring data is characterized by storing data in a compressed format with a high speed of reading and writing data, the possibility of scaling and expanding, a high level of reliability, fault tolerance and performance.

The control room serves to remotely provide processed data to the user through visualization tools as a result of monitoring and forecasting, and also displays information about recommendations for production processes and the causes of equipment failures.

The telemetry database allows you to store large amounts of data on the functioning parameters of transport nodes and systems for long periods of time.

The database of historical data on breakdowns is the initial and historical information about the functioning of the components, assemblies and systems of the vehicle, statistics on the causes of failures, repairs and maintenance. This allows, based on historical experience, to more accurately predict the condition of the vehicle.

The database of malfunctions and recommendations provides an opportunity to carry out long-term storage of the results of the system, as well as promptly transfer data to the driver and dispatcher. In addition, the dispatcher can request data from the database for a specific date or time period.

The vehicle condition monitoring unit operates on the basis of a trigger block, which allows taking into account not only individual parameters, but their cumulative change, which may be a sign of incorrect operation of the vehicle’s components and systems, and may also cause a risk of vehicle failure on the route and fatal consequences. .

The block of predictive analytics based on machine learning allows for complex prediction of the state of vehicle components and systems due to a large number of predictive analytics models. The use of machine learning allows forecasting with high accuracy.

Predictive analytics models allow you to predict the remaining resource of the node, predict the probability of failure in a specified time interval, perform anomaly detection in real time, determine the probability of failure by operating time or by mileage. In addition, the use of boosting technology makes it possible to determine model parameters (weight characteristics, parameters of variable relationships, etc.) as efficiently as possible at the lowest cost by using the experience of previous steps and the ability to check a large number of parameters in one iteration.

The sub-block of prediction and forecasting based on the model implements the formation of forecasts based on predictive strategies, i.e. identifying trends in parameter changes at different time intervals, abnormal deviations from trends, and determining the causes of anomalies based on specified correlations and historical data.

The subblock for interpreting the results and recommendations for decision-making generates recommendations for the decision-making process on the continuation of operation or the implementation of repairs necessary to eliminate the identified malfunctions.

The main purpose of the on-board analytical complex for vehicles is remote diagnostics and real-time monitoring of the technical condition of the fleet of vehicles (both unmanned and under the control of a human driver), as well as expert support for the decision-making process to continue operation or repair, in including the nature and extent of the repair necessary to eliminate the identified malfunctions. Due to its functional features, the platform allows monitoring the technical condition aimed at preventing critical transport failures and promptly responding if they occur, as well as predicting changes in the technical condition that can affect the performance of vehicle components, assemblies and systems.

The claimed complex contains:

1) Sensor for receiving the first type of data from the vehicle (telemetry data);

2) Sensor for receiving the second type of data from the vehicle (telemetry data);

3) Onboard analytical unit;

4) Sensor pre-processing sub-block 1;

5) Sensor pre-processing sub-block 2;

6) Subblock of temporary storage and transmission of data;

7) Subblock for assessing the actual performance of the vehicle;

8) Driver monitor;

9) Control room;

10) Block of data storage and structuring;

11) Telemetry database;

12) Base of historical data on breakdowns;

13) Database of faults and recommendations;

14) Unit for monitoring the state of the vehicle;

15) Data validation subblock;

16) Data processing sub-block;

17) Anomaly detection subblock;

18) Sub-block of scheduling for checking triggers;

19) Trigger subblock;

20) Block of predictive analytics based on machine learning;

21) Sub-block of the schedule for the functioning of predictive models;

22) N-number of predictive analytics models;

23) N-number of sub-blocks of primary data processing;

24) N-number of data extraction sub-blocks;

25) N-number of data transformation subblocks;

26) N-number of data transmission sub-blocks;

27) N-number of sub-blocks of feature extraction and data transformation;

28) N-number of model-based prediction and prediction sub-blocks;

29) N-number of sub-blocks of interpretation of results and recommendations for decision-making.

In FIG. 1 shows a diagram of the proposed device of the on-board analytical complex for vehicles.

The vehicle is equipped with two sensors 1 and 2 for obtaining different types of vehicle operation parameters data - telemetry data. The values of these parameters are fed to the onboard analytical unit 3.

The on-board analytical unit consists of the following sub-blocks: 4 - sub-block for preliminary processing of sensor data 1, 5 - sub-block for preliminary processing of sensor data 2, sub-block for temporary storage and transmission of data 6, sub-block for assessing the actual performance of the vehicle 7.

The data pre-processing sub-block 4 is used to receive data from the sensor 1, data deduplication, converting them into the format required for reading by the next block 5, and transmitting the processed data with data filtering. Data filtering allows, based on the built-in logic, to change the time intervals for collecting parameters and set the parameter Δ, which characterizes the amount by which the parameter must change for its registration. This is necessary in order to use resources as efficiently as possible and not overload them with large amounts of data. The data pre-processing sub-block 5 operates in a similar way to receive and process data from the sensor 2.

Data from data pre-processing sub-blocks 4 and 5 are transferred to the temporary storage and data transfer sub-block 6, which is used to transfer data to the data storage and structuring block 10, namely to the telemetry database 11 and to the sub-block for evaluating the actual vehicle performance 7. Temporary storage data is necessary in situations where the data storage and structuring unit 10 is not available for some reason or communication with it is lost.

Data from the sub-block of temporary storage and data transmission 6 is sent to the sub-block for evaluating the actual performance of the vehicle 7, which carries out the process of correlating the received data with triggers to identify violations in the operation of the subsystems and components of the vehicle, as well as generating alerts. Triggers take into account a limited number of parameters (telemetry data), an individual and cumulative change of which above (or below) certain values may be a sign of incorrect operation of the vehicle (assemblies and systems), and may also cause the risk of vehicle failure on the route and fatal consequences. The data from the sub-unit for assessing the actual performance of the vehicle 7 is sent to the driver monitor 8 and to the control room 9. The driver monitor 8 is installed in the driver's cab of the vehicle and allows you to quickly display limited information about violations in the operation of the systems and components of the vehicle and recommendations for making further decisions from the subblock for evaluating the actual performance of the vehicle 7 and the database of malfunctions and recommendations 13. The control room 9 allows you to remotely receive complete information from the subblock for assessing the actual performance of the vehicle 7 and the database of faults and recommendations 13, as well as data from the database of historical data on breakdowns 12 Also, control station 9 allows you to request data from the above modules for certain periods of time.

The data storage and structuring unit 10 consists of a telemetry database 11, which receives vehicle telemetry data (functional parameters of vehicle components and systems) from a temporary storage and data transmission sub-unit 5, a historical breakdown data base 12, in which data is stored and structured on the timing and scope of maintenance from various operating and maintenance organizations, and a database of faults and recommendations 13.

The data from the telemetry database 11 is sent to the vehicle condition monitoring unit 14, namely, to the data validation sub-unit 15, which checks the received data for their correctness. This sub-block checks whether the received data of any parameter corresponds to the data type assigned to it. Then the data from the sub-block 15 is sent to the data processing sub-block 16, which is designed to convert the data into the form necessary for the operation of the anomaly detection sub-block 17. An example of such a transformation is obtaining the parameter values necessary for the operation of certain triggers; removal of null values; selection of parameter values corresponding to certain time intervals. Data from subblock 16 is transmitted to anomaly detection subblock 17 - this is the process of correlating the received data with the trigger subblock 19 to detect violations in the operation of vehicle subsystems and components, as well as generating alerts. This sub-block operates according to the sub-block of generating a schedule for checking triggers 18, which sets the time intervals through which certain triggers should be checked. Trigger subblock 19 consists of N triggers 19.1-19.N and sets the parameters for the operation of subblock 17. Triggers take into account the individual and cumulative number of parameters (telemetry data), a change in which above (or below) certain values may be a sign of incorrect operation of the vehicle (nodes and systems) and may also cause the risk of vehicle failure. As a result of the operation of the vehicle condition monitoring unit 14, alerts are generated that are transmitted to the database of faults and recommendations 13.

In contrast to the sub-block for evaluating the actual performance of the vehicle 7, which operates in real time, the monitoring block for the condition of the vehicle 14 performs longer data processing, as it allows you to determine more complex triggers that require the values of a larger number of parameters and / or data to function. for long periods of time.

Data from the telemetry database 11 is also transmitted to a predictive analytics unit 20 based on machine learning, which operates through N number of models 22.1-22.N containing predictive analytics hypotheses and strategies. The sequence of data processing and analysis steps within each model 21.1-22.N will be the same, so the models consist of similar sub-blocks. The models work according to the data of the sub-block of the schedule for the operation of predictive models 21, which is a tool for performing tasks on a schedule: it forms a sequence of tasks, monitors their execution (in case of failure of any task, it gives information about an error), forms time intervals for starting tasks.

Each model contains a primary data processing sub-block 23.1-23.N, which includes a data extraction sub-block 24.1-24.N, a data conversion sub-block 25.1-25.N, a data transmission sub-block 26.1-26.N. The data extraction sub-block 24.1-24.N is required to obtain data from the telemetry database 12. Then the data is fed to the data transformation sub-block 25.1-25.N to extract the features necessary to use predictive models and transform the data into a form suitable for use by the model (these are algorithms for dimensionality reduction and vectorization of the state of transport over a significant period of time). The next data transfer subblock 26.1-26.N is responsible for transferring data that has undergone primary processing as a result of the operation of the module 23.1-23.N to the feature extraction and data transformation subblock 27.1-27.N.

Sub-block 27.1-27.N considers the statistical parameters of the predictive models, for example, the standard deviation. Then the data is fed to the prediction and forecasting sub-block based on the model 28.1-28.N, which serves to generate forecasts based on predictive strategies, i.e. identifying trends in parameter changes at different time intervals, abnormal deviations from trends, and determining the causes of anomalies based on specified correlations and historical data. This sub-block functions due to predictive analytics strategies that form predictive models and predictive data.

Strategy 1. Linear regression model for predicting the remaining resource of the node - allows you to determine the period (N days / cycles) of work until the node fails.

Strategy 2. Classification model for predicting the probability of failure in a specified time interval - allows you to determine if the node will break in the next N days / cycles of use.

Strategy 3. Real-time anomaly detection - allows you to determine the presence of anomalous behavior of indicators at the current time.

Strategy 4. A model that predicts the probability of failure by operating time or by mileage (Survival Analysis) - allows you to determine, based on the static characteristics of the node, the change in the risk of failure over time of operation.

The construction of predictive models is based on hypotheses that determine the interdependence of parameters and nodes, as well as the conditions by which it is possible to determine or predict the failure of a particular node (device).

In the subblock for interpreting the results and recommendations for making decisions 29.1-29.N, based on the work of the previous subblock 28.1-28.N, recommendations are formed for the decision-making process to continue operation or carry out repairs necessary to eliminate the identified malfunctions. The data obtained as a result of the operation of the block of predictive analytics based on machine learning 20 is transferred to the database of faults and recommendations 13. The database 13 allows you to store and structure the data of the results of the blocks 14 and 20.

An example of the invention is shown below.

Consider an example of the invention on a vehicle - a tram.

By using the sensor to receive the first type of data from the vehicle (telemetry data) (1), we can receive information such as:

– speed, km/h;

- cabin temperature (for sections 1 and 2), ºС;

– energy consumption (for 1+2 sections, 3 sections), kW/h;

– energy recovery (for 1+2 sections, 3 sections), kW/h;

– sand level in the sandbox, % (for 6 sandboxes);

– safety pedal, on / off;

- reverser (forward), on / off;

- reverser (back), on / off;

- overheating of brake rheostats (TR), presence / absence;

– connection with current inverter (IT), presence/absence (for 3 IT);

– emergency braking, on / off;

– excess voltage of the current inverter (IT), Volt (for 3 IT);

- excess of the alternating (Icd) current of the current inverter (IT), Ampere (for 3 IT);

- excess current of the phases of the current inverter (IT), Ampere (for 3 IT);

– failure of the current inverter driver (IT), presence / absence (for 3 IT);

– door anti-clamp, presence / absence / no power supply (for 6 doors);

– enable sandboxes, enabled / disabled;

– outdoor air temperature, ºС;

– brake, release / brake (for 6 brakes);

– “arrow”, on / off;

- "passage", on / off;

– section mains voltage, Volt (for 3 sections);

– section motor current, Ampere (for 3 sections);

- voltage of the battery (battery) of the section, Volt (for 8 batteries);

– battery current (battery), Amperes (for 2 batteries);

– rail brake, on / off;

– “air conditioner” button, on / off;

– emergency braking, on / off;

– emergency braking, on / off;

– deviation from the normal operating mode of the on-board converter (BP) (for 5 BP);

– travel / brake, on / off;

– travel/brake controller, on/off;

– connection of the three-phase on-board converter (BCT) (for 2 BCT);

– temperature of the current inverter (IT), ºС (for 3 IT).

By using the sensor to receive the second type of data from the vehicle (telemetry data) (2), we can receive information such as:

- type of transport;

- version of the program;

- cabin temperature, °С;

- outdoor illumination, Lum.;

- outdoor temperature, °С;

- reference temperature in the driver's cab, °С;

- reference temperature in the tram cabin, °С;

- temperature in the cabin, °С (for 3 sections);

- actuation of anti-clamp sensors at the doors (for 6 doors);

- door lock button, on / off;

- travel lock button, on / off;

- the presence of high voltage;

- car power supply, on / off;

- cabin heating button, on / off;

- button for additional cabin heater, on / off;

- interior heating button, on / off;

- glass heating button, on / off;

- button of the block of switching switches (BKV), on / off;

- automatic button 550/24, on / off;

- air conditioner button, on / off;

- automatic 550/24, on / off (for 2 sections);

- automatic cabin heating, on / off (for 2 sections);

- automatic interior heating, on / off (for 2 sections);

- automatic air conditioner, on / off (for 2 sections);

- autonomous running mode (AX) of the trolley, on / off;

- trolley in the high-voltage automatic (AV) mode from the local switch of the block of switching switches (LK-BKV), on / off (for 3 trolleys);

- trolley in autonomous running mode (AX) from the local switch of the block of switching switches (LK-BKV), on / off (for 3 trolleys);

- state of the autonomous running switch (AX) from the local battery switch (LC-ACB), on / off (for 2 AX switches);

- overheating of brake rheostats (TR), presence / absence (for 3 bogies);

- direction of movement, degrees;

- decrease in accuracy in the horizontal plane, meters;

- approximate error in determining the coordinates, meters;

- correctness of coordinates determination data;

- latitude and longitude, degrees;

- number of used satellites, pcs.;

- speed km/h;

- current date/time in UTC, hour/minute,

- parameters of 16 cameras in the cabin: camera name, connection with the camera, camera IP address, number of camera failures;

- the number of hard drive failures since power on, the number of times;

- readiness of the hard drive;

- filling of the write buffer, %;

- the name of the hard disk model;

- number of write buffer overflows since switching on, number of times;

- state of the hard drive;

- hard disk temperature, °C.

The values of these parameters are sent to the onboard analytical unit (3) via secure VPN channels using Ethernet. The onboard analytical unit is a compact complex for high-precision navigation and high-speed mobile communication, which is designed to aggregate communication channels into a single broadband channel and transmit data through several independent channels, including those belonging to different telecom operators. With the help of aggregation of communication channels, the problem of stability / reliability and the maximum speed of transmitted data is solved.

The on-board analytical unit also contains sensor data pre-processing sub-units 1 and 2 (4, 5) - these are software solutions that make a request for data and process it.

The software components of the onboard analytical unit (3) include a temporary data storage and transmission subunit (6) - a proxy server that serves to cache and transfer data to the data storage and structuring unit deployed on the server base. Data caching is necessary in situations when the server is not available for some reason or communication with it is lost, and allows temporary storage of data. The data processed by the onboard analytical unit via communication elements - antenna and communication unit (3G / LTE) are received in near real time mode (Near Real-Time) to the data storage and structuring unit (10) through secure VPN channels using the HTTP protocol.

The sub-block of assessing the actual performance of the vehicle (7) is a software solution based on triggers. The list of triggers includes the types of notifications corresponding to one or another trigger: EO - emergency notification (high criticality level); O - alert (medium criticality level); P - warning (informing), as well as a software description of the operation process, based on the telemetry parameters received from the trams.

The driver's monitor (8) can be made using a touch screen and is intended for driver / driver interaction with the software and hardware complex and displaying the necessary parameters.

The driver's monitor is used to control the digital information system, display diagnostic and other information. As a display, it is better to use matrices with high resolution, wide viewing angles and high brightness.

Control room (9) - a web interface that serves to provide processed data to the user through visualization tools, such as Grafana. This service displays vehicle operation parameters, predictive analytics and monitoring data. Separately organized access to the historical database of breakdowns.

All the items described below work on the basis of the server and represent software-algorithmic data processing.

Telemetry database (11) - a solution that uses the Clickhouse system for data storage. ClickHouse is an open-source column-based analytical database management system (DBMS) that allows you to run analytical queries in near real-time mode (Near Real-Time). This solution allows you to store data for a long period of time. Clickhouse generates data requests and organizes them by indexing by host (tram number), by signal type (parameter name), and by time. This indexing allows you to quickly get the necessary data from the database. As a result, a table is formed that shows the parameters received from a particular tram: parameter fixation time; parameter name, parameter value and assignment of this parameter to a specific data type (float or int).

The database of historical data (12) about breakdowns can work on the basis of the PostgreSQL database management system and contains data on the terms and volumes of maintenance from various operating and maintenance organizations - in the form of tables. This data contains such important, within the framework of predictive analytics, information as:

- side and inventory number of the vehicle;

- date of detection of the malfunction;

- date of troubleshooting

- circumstances for detecting a malfunction: return, EO (daily maintenance), RK (control inspection), TO (technical maintenance);

- type of fault, description;

- the name of the node in which the malfunction was detected;

- node area: braking system, traction, climate system, interior access, etc.

- mileage of the vehicle at the time of detection of the malfunction;

- place of registration of the vehicle.

The fault and recommendations database (13) works similarly to the telemetry database through the use of Clickhouse and stores the results of the system - monitoring and predictive analytics.

The monitoring block (14) receives the data from the telemetry database running on the basis of Clickhouse and processes them according to certain rules - triggers.

Data validation subblock (15) - checking the received data for their correctness. The parameters used by the monitoring block can have the following three data types: float, int, string. This sub-block checks whether the received data of any parameter corresponds to the data type assigned to it. String is a text data format, in our case it is the name of the parameter (the name of the parameter under which the data is stored in the system); int - parameters that have an integer value; float - floating point parameters.

Data processing subblock (16) - converting data into the form necessary for the functioning of triggers. An example of such processing would be getting parameter values required for certain triggers to work; removal of null values; selection of parameter values corresponding to certain time intervals.

Anomaly detection subblock (17) - the process of correlating the received data with triggers to detect violations in the operation of vehicle subsystems and components, as well as generating alerts according to the scheduling subblock for checking triggers (18).

The anomaly detection sub-block operates due to the data of the trigger sub-block (19), which take into account a separate or cumulative number of parameters, the change of which above (or below) certain values may be a sign of incorrect operation of the vehicle (assemblies and systems). For example, if a tram has a speed of more than 5 km / h and at the same time the hand brake is on, this is a trigger. An example of another trigger is when the air conditioner is turned off for more than 25 minutes when the cabin temperature is over 25 degrees Celsius. Triggers are some rules that identify violations in the operation of the vehicle. Thus, this software-algorithmic solution includes data on the permissible values of parameters and their sets, as well as the logic for generating alerts based on these data corresponding to a particular trigger, for example: EO - emergency alert (high criticality level); O - alert (medium criticality level); P - warning (informing).

The block of predictive analytics based on machine learning (20) is designed to prevent critical transport failures, as well as to predict changes in the technical condition of the vehicle that can affect the performance of vehicle components, assemblies and systems. This block functions on the basis of the subblock of the schedule for the operation of predictive models (22.1-22.N), which is a software solution.

The predictive analytics block works through models that contain hypotheses and predictive analytics strategies.

For the models to work, a number of hypotheses must be formulated - assumptions about what parameters and conditions can be used to determine or predict the breakdown of a particular node (device).

For example, for a tram, hypotheses can be identified for the following nodes:

- traction motor inverter;

- actuating mechanisms of brakes;

- doors;

- speed sensor (speed sensor);

- high-voltage high-speed automatic machine (VB-11);

- motor traction inverter fan;

- brake rheostat.

An example of a hypothesis for a motor traction inverter: it is possible to determine the breakdown of the motor traction inverter by the difference in motor current parameters, as well as predict the breakdown of the motor traction inverter by telemetry of motor currents, reaching an accuracy (recall) of 0.8, with a sensitivity of at least 0.8 (Recall shows what proportion objects of a positive class from all objects of a positive class found the algorithm).

N-number of primary data processing subblocks (23.1-23.N) is responsible for extracting, transforming and loading data.

N-number of data extraction subblocks (24.1-24.N) means loading data from a telemetry database running on Clickhouse. Thus, a table is formed containing the date and time when the telemetry parameter was received, its name (name) and the parameter value itself, referred to value_float (floating point value) and value_int (integer values) depending on the data type.

N-number of data transformation sub-blocks (25.1-25.N) - extraction of features necessary to use predictive models and transformation of data into a form suitable for use by the model (these are algorithms for dimensionality reduction and vectorization of the state of transport over a significant period of time).

As an example of this step, a software algorithm for the operation of the brake actuator hypothesis (BMI) shows the conversion of BMI actuation data into time delays between actuation signals. As a result, a data set is formed, which includes: delay list - a set of all delays on the axes for a given period of time, delay_daily_pct - a list containing average delay values for each day; average delay - average value for all delays; min_date and max_date - time interval for receiving parameters; delay_daily_pct average - average delay calculated relative to the data set from column 3; 1-6_axis, % - percentage of the number of delays on a specific axis to the total number of delays, not null IMT values - necessary parameters for the model to function.

N-number of feature extraction and data transformation subblocks (27.1-27.N) is the unit that counts the statistical parameters of the predictive models, such as standard deviation.

N-number of sub-blocks of prediction and forecasting based on the model (28.1-28.N) - the formation of forecasts based on predictive strategies, i.e. identifying trends in parameter changes at different time intervals, abnormal deviations from trends, and determining the causes of anomalies based on specified correlations and historical data.

This module will function through predictive analytics strategies that generate predictive models and predictive data. The construction of predictive models is based on hypotheses that determine the interdependence of parameters and nodes, as well as the conditions by which it is possible to determine or predict the failure of a particular node (device).

Strategy 1. Linear regression model for predicting the remaining resource of the node - allows you to determine the period (N days / cycles) of work until the node fails.

In it x ₁ is the value of some telemetry parameter at time t ₁ , x _n is the value of the same parameter at time t _n . The task of this strategy is to determine the optimal values of the coefficients of the regression equation w ₁ …w _n , at which the value of the loss function will be minimal. Thus, the function f ( x ) is formed - the equation of a straight line that describes the behavior of the parameters over time. Based on this, it is possible to predict the value of the parameter x at some of the following times - t _{n +1} ... t _{n + k} .

This strategy allows you to determine the period (N days / cycles) of operation until the node fails in cases where the node resource is limited and has some value. For example, when it is known that when a certain mileage is reached, it is highly likely that a part will need to be replaced. Using this strategy, you can determine how long a given mileage will be reached in the future.

Also, the strategy can be used as an initial step, which allows predicting the behavior of a parameter over time, for subsequent classification models.

Strategy 2. Classification model for predicting the probability of failure in a specified time interval - allows you to determine if the node will break in the next N days / cycles of use.

As part of this predictive analytics strategy, a binary sign will be used that answers positively and negatively to the question: will the node break in the next N days / cycles of use; 1 - yes, it will break, 0 - no, it will not break.

Thus, X→D _f is the set of admissible values of the attribute (ie 1 or 0); X is a matrix of values for the functioning parameters of certain nodes required for a detuned hypothesis, Y is a class to which a certain set of parameters belongs. For example, it is known that when the current inverter reads more than 300 A ( x ₁ ) and the section motor current reads more than 600 A ( x ₂ ), the node is broken ( y ₁ ). Thanks to this hypothesis, it is possible to determine whether the node ( D _f ) will break with other values of this parameter. This example is conditional, in practice, to determine D _f , it is necessary to have large training samples - y values for x parameters and a well-trained classification model.

Strategy 3. Real-time anomaly detection - allows you to determine the presence of anomalous behavior of indicators at the current time.

In data analysis, there are two sections that are aimed at finding anomalies: outlier detection (Outlier Detection) and "novelty" (Novelty Detection).

Outliers can appear as a result of:

- data errors (measurement inaccuracies, rounding off, incorrect data recording, etc.);

- the presence of noise objects (ie objects that are incorrectly classified);

- the presence of objects from "other" samples (this can be caused, for example, by readings from a broken sensor).

Novelty most often appears due to a fundamentally new behavior of the object, for example, the operation of an engine after a breakdown. Within the framework of predictive analytics, what is interesting is the search for novelty to determine the breakdowns of devices and nodes.

Strategy 4. A model that predicts the probability of failure by operating time or by mileage (Survival Analysis) - allows you to determine, based on the static characteristics of the node, the change in the risk of failure over time of operation.

To search for optimal algorithms for a particular strategy with respect to a certain hypothesis, experiments are carried out to evaluate the effectiveness of these algorithms. In addition, ensembles can be used to find optimal mathematical models.

An ensemble is a set of predictors that together give the best result, in contrast to a single predictor. Ensembling technologies are further divided into Bagging and Boosting. Boosting is most effective for the classic machine learning algorithms described above, which are used as part of predictive analytics.

Boosting allows you to determine hyperparameters as efficiently as possible at the lowest cost - the parameters of a mathematical model (weight characteristics, parameters of variable relationships, etc.) by using the experience of previous steps and the ability to check a large number of hyperparameters in one iteration.

Using the strategies and algorithms described above, as well as boosting technology, it is possible to develop predictive models for the operation of the N-number of model-based prediction and prediction sub-blocks (28.1-28.N).

Interpretation of results and recommendations for decision-making (29.1-29.N) - based on the work of the previous module, recommendations are formed for the decision-making process on the continuation of operation or the implementation of repairs necessary to eliminate the identified malfunctions.

Claims (4)

1. An on-board analytical complex for vehicles, containing a sub-block of permissible, nominal, limiting values of parameters, a sensor for obtaining data from a vehicle, a sub-block for preliminary processing of sensor data for obtaining data from a vehicle, a sub-block for assessing the actual performance of the vehicle, an information and control panel , a database of faults and recommendations, a timer block, characterized in that a second sensor is additionally introduced for receiving data from the vehicle and a second sub-block for pre-processing data of the second sensor for receiving data from the vehicle, while the sub-block for assessing the actual performance of the vehicle is included in the additionally introduced on-board analytical unit, which also includes a temporary storage and data transfer unit, the driver’s monitor is used as an information and control panel, while the database of malfunctions and recommendations cations is included in the additionally introduced block for storing and structuring data, while the complex additionally contains a block of predictive analytics based on machine learning, a block for monitoring the state of the vehicle, which additionally includes a subblock of triggers that performs the functions of a subblock of permissible, nominal, limiting values of parameters, and a subblock for generating schedules for checking triggers, which performs the functions of a timer block. 2. An on-board analytical complex for vehicles according to claim 1, characterized in that the block for storing and structuring data is equipped with a telemetry database and a database of historical data on breakdowns. 3. An on-board analytical complex for vehicles according to claim 1, characterized in that it is equipped with a control room, which is a web interface. 4. An on-board analytical complex for vehicles according to claim 1, characterized in that the predictive analytics block based on machine learning contains a sub-block of the schedule for the operation of predictive models, which acts as a timer block, and N is the number of predictive analytics models.

Images