WO2013037594A1 - Method and system for detecting defects in an air-cooled condenser used in a vapor-compression cycle machine - Google Patents

Abstract

The invention relates to a method and a system for detecting defects in an air-cooled condenser used in a vapor-compression cycle machine. The method involves determining the existence of defects in the air-cooled condenser on the basis of only six parameters, i.e. the condensation temperature (T°cond) of the air-cooled condenser (Cond), the outside temperature (T°ext) of the machine, the temperature (T°fc) of the heat-transport fluid (FC), the machine-overheating value (VS), the compressor control (CC), and the fan (Vent) control (CV). The method makes it possible to determine a deviation indicator (Q) from said parameters and estimated values representative of the normal operation of the machine for said operating point of the machine, said deviation indicator being compared with a predetermined threshold value (δ) in order to deduce whether or not a defect exists.

Description

 Method and system for detecting defects in an air condenser used in a vapor compression cycle machine Technical field of the invention

 The invention relates to the detection of defects in an air condenser employed in a steam compression cycle machine for use in the HVAC (Heating, Ventilation and Air Conditioning) field (or heating, ventilation and air conditioning).

State of the art

 There are many documents proposing methods for detecting abnormal situations in heat exchangers. This is the case for example in the documents US786621 1, US201 1 144807 or US2010102828. However, the solutions proposed in these documents require the installation of additional and specific sensors on the machine to perform the diagnostics. Other solutions such as those described in US6386272 or US200513321 1 must take into account a physical model of the machine to be able to make a diagnosis. Document US2007 / 089438 describes a method for controlling a refrigerant in a refrigeration system.

 The object of the invention is to propose a method for detecting defects in an air condenser of a vapor compression cycle machine, which is simple and reliable and which can operate without the installation of additional sensors on the machine. , and without taking into account a physical model of the machine.

Presentation of the invention

This object is achieved by a defect detection method in an air condenser used in a steam compression cycle machine, said machine comprising at least one compressor connected to the air condenser and intended to ensure the circulation of a refrigerant an evaporator connected to the air condenser via an expander and traversed by a coolant, said air condenser comprising at least one fan, the method comprising steps of: obtaining several parameters at a given instant for an operating point of the machine, said parameters being the condensation temperature of the air condenser, the external temperature of the machine, the temperature of the coolant, the value of the superheat of the machine, machine, compressor control and fan control,

determining an indicator of deviation between the parameters obtained and estimated values, said estimated values representing the normal operation of the machine for said operating point,

 - determination of defects by comparison of said deviation indicator with respect to a determined threshold value.

 According to a particularity, the method comprises a learning step consisting of:

 determining an operating range of the machine having operating points,

 - record said parameters for several operating points of the operating domain.

 In another feature, the operating domain is divided into classes and a class is considered learned if enough operating points of the class have been observed.

 According to another particularity, the estimated values are determined from a statistical model.

 According to another particularity, the statistical model is calculated on the operating domain, once enough operating points have been observed during the learning step.

 According to another particularity, the statistical model is obtained by a principal component analysis.

 According to another particularity, the threshold value is determined by statistical study of the deviation indicator.

According to another particularity, the step of determining the fault consists in analyzing the control of each fan. According to another particularity, the value of superheating is determined from the temperature of the refrigerant sucked by the compressor and the pressure of the refrigerant at the outlet of the evaporator.

 The invention also relates to a system for detecting defects in an air condenser of a steam compression cycle machine. This system comprises:

 means for obtaining several parameters at a given moment for an operating point of the machine, said parameters being the condensation temperature of the air condenser, the external temperature of the machine, the temperature of the heat transfer fluid, the value of overheating of the machine, compressor control and fan control,

 means for determining an indicator of deviation between the parameters obtained and estimated values, said estimated values representing the normal operation of the machine for said operating point,

- Fault determination means by comparing said deviation indicator with respect to a determined threshold value.

 According to another feature, the system comprises means for determining a statistical model for calculating the estimated values.

Brief description of the figures

 Other features and advantages will appear in the following detailed description with reference to the accompanying drawings in which:

 FIG. 1 schematically represents a steam compression cycle machine,

 FIG. 2 represents a functional diagram explaining the fault detection method of the invention,

FIG. 3 represents an operating range of the vapor compression cycle machine and illustrates the principle of the learning method used for the method of the invention, FIGS. 4A and 4B illustrate the fault detection principle employed in the method of the invention.

Detailed description of at least one embodiment

 The invention relates to the detection of defects in a vapor compression cycle machine such as can be employed for HVAC systems.

 In a known manner a vapor compression cycle machine such as that shown in FIG. 1 mainly comprises:

 - A condenser with air Cond allowing to pass a refrigerant

Ff from the gaseous state to the liquid state. The air condenser Cond can be for example of the tube and fin or microchannel type and may include one or more fans Vent to ensure an air passage through the air condenser Cond to allow the condensation of fluid refrigerant Ff.

 A pressure regulator Det making it possible to reduce the pressure of the refrigerant Ff.

 An Ev evaporator for passing the refrigerant Ff from the gas + liquid state to the gaseous state. The evaporator Ει / is also traversed by a heat transfer fluid Fc, for example air or glycol water, exchanging its calories with the refrigerant Ff to be heated or cooled.

 One or more compressors Comp intended to suck the refrigerant Ff in the gaseous state from the evaporator Ev to send it to the air condenser Cond.

The machine also has several sensors used for controlling the machine. The invention has the advantage of not requiring the addition of additional sensors. The sensors already present on the machine are sufficient for the implementation of the invention. The machine comprises an external temperature sensor (7 ^~ ¾x /), a sensor of the suction temperature (7 ^~ ¾uc) of the compressor Comp, a heat transfer fluid temperature sensor (T ° fc), a sensor of pressure (LP) of the refrigerant at the outlet of the evaporator Ev and a pressure sensor (HP) of the refrigerant at the inlet of the air condenser Cond.

The invention consists in implementing a method for detecting defects in the Cond air condenser used in a steam compression cycle machine as described above. This method is implemented by a detection system that can be integrated into the computer already present on the machine or be external to it.

The detection method has the particularity of requiring only the use of six simple parameters already available because necessary to control the machine. These six parameters are as follows:

- The condensing temperature 7 ^" ¾ond in the Cond air condenser,

the external temperature 7 ^~ ¾xf to the vapor compression cycle machine, that is to say the ambient temperature,

 the temperature of the coolant T ° fc at the outlet or at the inlet of the evaporator Ev,

 the superheat value VS of the machine, which can be obtained directly or by calculation from the temperature of the refrigerant sucked by the compressor Comp and from the pressure of the refrigerant Ff at the outlet of the evaporator Ev,

 the control of the DC compressors, which can be in all or nothing or correspond to a variable controlled by a variable speed drive,

- The control of the fans CV, it can be all or nothing or correspond to a variable controlled by a variable speed drive.

With reference to FIG. 2, the determination of the defects in the air condenser Cond at a given moment and for a given operating point of the machine consists first of all in determining a distance indicator Q from the six parameters. defined above at the given time and at the point of operation and to compare the indicator of difference Q obtained with a threshold value ^ predetermined to deduce the presence or absence of a defect and possibly the nature of said defect. The determination of the difference indicator Q involves the creation of an MS statistical model (FIG. 2) making it possible to determine estimated values for each operating point of the machine, said estimated values thus representing the normal operation of the machine. at each operating point.

 The statistical model MS is created after a learning step of the machine. The learning step consists of observing the different operating points of the machine and recording said six parameters defined above for each of the operating points. The statistical model MS is generated once enough operating points have been observed. The statistical model MS thus represents the normal operation of the machine on a particular operating domain.

 The learning step can be triggered in different ways:

 - manually by an operator,

 - automatically when the system detects that the characteristics of the machine have returned to normal or when an analysis of the characteristics of the machine shows that they are too far from normal and that it is necessary to take into account aging of the machine.

 According to the invention, it is not necessary to wait for the learning to be complete, that is to say that all operating points have been scanned before being able to begin to detect faults in the condenser. Air Cond. Indeed, a complete learning requires being able to scan all operating points of the machine, but this is not feasible over a short period because the machine does not have time to be confronted with different climatic conditions.

Therefore, the invention consists in being able to carry out the detection of defects even if the learning step is not completely finished. For this, we define an operating range of the machine that is divided into several classes C1, C2, C3, Cn (hereinafter C1-Cn), each class can be learned independently of another. If the machine is on an operating point included in a class already learned, fault detection will be possible for this operating point. According to the invention, the learning step consists first of all in determining the operating range of the machine. The operating domain Df of the machine is defined by several of the six parameters and makes it possible to implement a learning step of the machine, this learning step being necessary to perform a fault detection in the air condenser Cond. . The operating domain Df is for example defined by two or three parameters.

 To define the operating range, it is necessary, for example, to determine the set temperature of the heat transfer fluid Fc and then to define the operating range around this setpoint temperature. This can also be achieved by using parameters present at the input of the air condenser fault detection method, by means of parameters known by default or by means of parameters determined by observation of the machine over a certain period of time.

In FIG. 3, for reasons of simplification, the operating range has for example been defined from two parameters, the outside temperature 7 ^~ ¾xf between -10 ° C. and + 40 ° C. and a compressor control between 0% and 100%. We thus have:

Df = [T ^ ext min; 7 ^~ ¾xf max] x [CC min; CC max]

 Of course, it would be possible to define another operating domain. In a three-dimensional operating range, the three parameters would be, for example, the external temperature T ° ext, the control of the compressors CC and the temperature of the coolant T ° fc. We would then obtain:

Df = [T ^ ext min; 7 ^~ ¾xf max] x [CC min; CC max] x [T ° fc min; T ° fc max]

With reference to FIG. 3, this operating domain thus delimited is divided into classes C1-Cn each corresponding to a zone of the operating domain. Each class C1-Cn may have several operating points. The operating domain is divided into one hundred classes. In FIG. 3, each class is a square delimited by a minimum value and a maximum value of the outside temperature and by a minimum value and a maximum control value of the compressors. In a three-dimensional functioning domain, each class would be represented by a cube.

During learning, the system constantly recovers the six parameters defined above and records the values for each point of operation of the machine. The system will consider that an operating point of the machine is observed if the six parameters have been recorded for this operating point.

 According to the invention, it can be considered that a class C1-Cn is learned when a certain number of operating points included in the class have been observed. For example, a class can be considered to be fully learned when at least one thousand operating points of the class have been observed and not learned when less than one thousand points of the class have been observed. However, it is also possible that a class, a priori not learned because not having enough observed operating points, is considered learned by extension if its neighboring classes are learned completely. For this we can define a convex envelope of fully learned classes in which the untrained classes become learned classes by extension. For each operating point of the machine included in a class learned completely or learned by extension, it will be possible to make the detection of air condenser defects, the statistical model MS being valid for said class. In addition, over time, a class learned by extension can be completed to eventually become a fully learned class. During the detection of a fault for an operating point of the machine, it will be possible to assign a confidence index taking into account the learning level of the class comprising this operating point.

 In addition, the state of each class can change over time. For example, the system is arranged to gradually eliminate the oldest observed points in order to take into account the aging of the machine, some fully learned classes may become untrained if they have lost one or more observed operating points. These lost operating points can be learned again during a new learning.

 In Figure 3, the fully learned classes are indicated in light gray, those that are by extension in dark gray and the parts indicated in intermediate gray allowed the construction of the convex form.

Once the operating domain is defined and enough operating points have been observed on the operating domain, the fault detection method is to determine the statistical model MS of the machine. This statistical model MS will be valid for classes learned completely or by extension. For example, the determination of the statistical model may begin when enough operating points have been observed.

 The statistical model MS of the machine is determined by principal component analysis (PCA) and can be used to determine estimated values for each operating point in a learned class. The statistical model is determined by the system as follows:

The system determines the variance-covariance matrix M of the six parameters defined above. It is fed by the parameters recorded for the various operating points. This variance-covariance matrix M is then diagonalised (for example by the Jacobi iterative method), which makes it possible to obtain a transition matrix R and a diagonal matrix D such that M = R * D * ^t R, the sign ^* being the sign of multiplication

 It is then necessary to determine the proper subspace that is the most important so as to obtain a matrix P deduced from R.

 Once these R and P matrices calculated by the system and defining the statistical model MS, the system can determine in real time a deviation indicator Q. As described above, the statistical model MS is only valid if the class C1-Cn which includes the point of operation of the machine that is observed, is considered learned by the system. Therefore, the difference indicator Q can only be determined for an operating point if it is in a learned class. The distance indicator Q is expressed by the following relation:

Q = norm (xP * ^t P * x) ² for x belonging to the operating domain in which:

x is the vector of the six parameters: x = (CC, CV, VS, T ° cond, T ° ext, Tic), norm is the norm of the vector xP * ^t P * x (for example, norm 2), fP is the transpose of the matrix P, - ² is the power function 2. Once the distance indicator Q is determined by means of the statistical model MS, the system performs a function Fdef (FIG. 2) for detecting defects of the air condenser Cond, which consists in making a comparison of the indicator with a value determined threshold. If the difference indicator Q is greater than the threshold value δ for a determined duration (for example T = 60s), the system detects a fault.

 The threshold value can be determined in two different ways:

 By automatic statistical study: the signal Q is observed by the system over a certain duration (for example a week) and the system determines its mean μ and its standard deviation (for example by an iterative method to reduce the memory space). The system then determines a confidence interval in which it is assumed that there are no defects. This confidence interval can be parameterized by the user or determined by assuming that the deviation indicator Q is Gaussian.

 - Considering that the input signal x (with six coordinates) is bounded, and using the eigenvalues of the matrix D, it is possible to determine an alarm threshold signaling the presence of a fault. This avoids waiting for a day of observation on the indicator Q. Once determined, the threshold value is stored by the system in storage means.

The comparison of the difference indicator Q with the threshold value δ makes it possible to detect the presence of a fault on the air condenser Cond. In addition, depending on the condition of each Vent fan, it is also possible to deduce if the fault results from a failure of a Vent fan or fouling of the Cond air condenser.

Indeed, if the deviation indicator Q is greater than the threshold value δ regardless of the Vent fans in operation, the system concludes that the condenser Cond. The Denc degree of fouling of the air condenser, expressed as a percentage, is then indicated to the operator. On the other hand, if the Q deviation indicator is greater than the threshold value δ only when one or more of the Vent fans are in operation, the system concludes that one or more of these Vent fans have failed. According to the invention, the determination of the defect can be accompanied by that of a confidence index IC (for example expressed as a percentage) making it possible to indicate to the operator a probability on the nature of the detected defect. This confidence index may for example depend on the state of the class, that is to say, learned or learned by extension, in which is the operating point. The confidence index IC will thus be higher if the operating point is in a fully learned class than if it is in a learned class by extension.

 Figures 4A and 4B illustrate the principle of fault detection. They represent the variation of the distance indicator Q as a function of time with respect to the threshold value as well as the state of the control of each fan Air condenser wind (ventilator activated indicated by a gray box and ventilator off signaled by a white box).

 In FIG. 4A, it can be seen that the difference indicator Q exceeds the threshold value for a long period, which causes the system to signal the presence of a fault D1.

 In FIG. 4B, the system notices that the Q indicator exceeds the threshold value over different periods during which the fan # 2 is operating. The system can therefore conclude that there is a fault (D2, D3, D4) when fan # 2 is supposed to work.

 In FIGS. 4A and 4B, it should be noted that if the difference indicator Q exceeds the threshold value δ for a period which is too short, the system does not take account of this temporary overshoot and does not conclude at the presence of a defect. .

Claims

1. Method for detecting defects in an air condenser used in a steam compression cycle machine, said machine comprising at least one compressor (Comp) connected to the air condenser (Cond) and intended to ensure the circulation of a refrigerant (F /), an evaporator (Ev) connected to the air condenser (Cond) via an expander (Det) and traversed by a coolant (Fc), said air condenser comprising at least one fan (Vent), characterized in that the process comprises steps of:  obtaining several parameters at a given instant for an operating point of the machine, said parameters being the condensation temperature (7 ~ ¾ond) of the air condenser (Cond), the external temperature (7 ~ ¾x /) of the machine , the temperature (T ° fc) of the coolant, the value of the superheat (VS) of the machine, the control (CC) of the compressor and the control (CV) of the fan, determining a deviation indicator (Q) between the parameters obtained and estimated values representing the normal operation of the machine for said operating point, said estimated values being derived from a statistical model (MS), said statistical model being determined by a principal component analysis after a machine learning step,  - Determining faults by comparing said deviation indicator (Q) with a determined threshold value (δ). 2. Method according to claim 1, characterized in that it comprises a learning step consisting of:  determining an operating range of the machine comprising operating points,  - record said parameters for several operating points of the operating domain. 3. Method according to claim 2, characterized in that the operating domain is divided into classes (C1-Cn) and in that a class (C1-Cn) is considered as learned if enough operating points of the class have been observed. 4. Method according to claim 2 or 3, characterized in that the statistical model (MS) is calculated on the operating domain, once enough operating points have been observed during the learning step. 5. Method according to claim 1, characterized in that the threshold value (δ) is determined by statistical study of the deviation indicator (Q).  6. Method according to claim 1, characterized in that, when the deviation indicator (Q) is greater than the threshold value (<5), the step of determining the defect comprises analyzing the control of each fan ( Wind).  7. Method according to claim 1, characterized in that the superheating value is determined from the suction temperature (7 ~ ¾uc) of the compressor and the pressure of the refrigerant at the outlet of the evaporator.  8. System for detecting defects in an air condenser used in a steam compression cycle machine, said machine comprising at least one compressor (Comp) connected to the air condenser (Cond) and intended to ensure the circulation of a refrigerant (Ff), an evaporator (Ev) connected to the air condenser (Cond) via an expander (Det) and traversed by a coolant (Fc), said air condenser comprising at least one fan (Vent), characterized in what the system includes:  means for obtaining several parameters at a given moment for an operating point of the machine, said parameters being the condensation temperature (7 ~ ¾ond) of the air condenser, the external temperature (7 ~ ¾x /) of the machine, coolant temperature (T ° fc), machine superheat value (VS), compressor control (CC) and fan control (CV), means for determining a difference indicator (Q) between the parameters obtained and estimated values representing the normal operation of the machine for said operating point, said estimated values being derived from a statistical model (MS), said statistical model being determined by a principal component analysis after a machine learning step,  - Fault determination means by comparing said deviation indicator (Q) with respect to a determined threshold value (δ).