A method and system for fault detection for a chiller

Abstract

The present application relates to the field of refrigerator fault detection, and more particularly to a fault detection method and system for a refrigerator, the method comprising: obtaining a multidimensional parameter set of any refrigerator in history, and calculating the recognition significance of any dimension; collecting a multidimensional parameter set of a refrigerator to be detected, constructing a target window of the refrigerator to be detected, segmenting the target window to obtain a plurality of data segments, performing PCA dimension reduction on the target segment to obtain principal components, projection value vectors and loading vectors, calculating a first discrimination distance according to the projection value vectors, and calculating a second discrimination distance according to the loading vectors and the recognition significance; calculating the recognition contribution degree of the target segment based on the first discrimination distance and the second discrimination distance; and calculating the fault degree of the refrigerator to be detected based on the recognition contribution degree and the second discrimination distance, thereby completing fault detection. The technical scheme of the present application can improve the accuracy of the fault detection result of the refrigerator.

Description

Technical Field

[0001] This invention relates to the field of refrigeration machine fault detection. In particular, it relates to a method and system for refrigeration machine fault detection. Background Technology

[0002] As the core component of refrigeration equipment, the reliability and stability of the refrigeration unit directly affect the performance and operating efficiency of the entire system. However, due to the long-term operation of the refrigeration unit under high load and high frequency working conditions, various faults occur frequently, leading to equipment instability, reduced efficiency, and even serious malfunctions, causing significant losses to production and use.

[0003] Most existing refrigeration equipment fault detection technologies rely on traditional manual inspection or fault detection using single-parameter or pairwise parameter coupling. For large-scale refrigeration equipment, manual inspection often fails to detect early faults in a timely manner, leading to prolonged operational abnormalities before serious failures occur, potentially causing complete system shutdown and equipment damage. Furthermore, fault detection using single-parameter or pairwise parameter coupling has an average fault detection cycle of 45–180 days, with 85% of micro-leakages and efficiency degradation going undetected until the compressor is completely damaged, resulting in inaccurate fault detection results. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.

[0005] In a first aspect, a fault detection method for a refrigeration machine includes: acquiring a multi-dimensional parameter set of any refrigeration machine in history, and calculating the recognition saliency of any dimension; collecting a multi-dimensional parameter set of the refrigeration machine to be detected, constructing a target window of the refrigeration machine to be detected, with the end of the target window being the detection time, segmenting the target window to obtain several data segments, taking any data segment as the target segment, performing PCA dimensionality reduction on the target segment to obtain principal components, projection value vectors, and loading vectors, with one principal component corresponding to one projection value vector and one loading vector, calculating a first discrimination distance based on the projection value vector, and calculating a second discrimination distance based on the loading vector and the recognition saliency; calculating the recognition contribution of the target segment based on the first discrimination distance and the second discrimination distance; and calculating the fault degree of the refrigeration machine to be detected based on the recognition contribution and the second discrimination distance, thereby completing the fault detection.

[0006] Preferably, the calculation of the recognition saliency of any dimension includes: performing PCA dimensionality reduction on the multi-dimensional parameter set of any refrigerator in history to obtain the eigenvalues ​​of each principal component and the loadings of each dimension; constructing a loading vector from the loadings of all dimensions in the same principal component; taking any dimension as the target dimension, calculating the product of the absolute value of the loading of the target dimension in any principal component and the eigenvalue of any principal component as the dimension weight of the target dimension; using the sum of the dimension weights of all principal components in the target dimension as the first saliency distance of the target dimension; taking the loading vectors corresponding to the principal components whose target dimension loadings are not equal to 0 as effective vectors; classifying refrigerators in history into normal refrigerators and abnormal refrigerators; taking the effective vector of any abnormal refrigerator as an abnormal effective vector; and taking any... The effective vector of a normal refrigeration unit is taken as the normal effective vector. The absolute value of the similarity between any abnormal effective vector and any normal effective vector is calculated. The sum of the absolute values ​​of the similarity between any abnormal effective vector and each normal effective vector is taken as the first accumulated value. The first accumulated value of each abnormal effective vector is obtained through iteration. The sum of all the first accumulated values ​​is taken as the second saliency distance of the target dimension. The absolute difference between the first saliency distance of any abnormal refrigeration unit and the first saliency distance of any normal refrigeration unit is calculated. The ratio of the absolute difference to the second saliency distance is calculated. The ratio of any abnormal refrigeration unit to each normal refrigeration unit is obtained through iteration. The sum of all the ratios is taken as the recognition saliency of the target dimension.

[0007] Preferably, the construction of the target window for the refrigerator to be tested includes: randomly dividing the multi-dimensional parameter set of the refrigerator to be tested into several initial windows of unequal length; for a segmentation result, calculating the DTW distance between any two window parameter sets, and calculating the average distance of all DTW distances; similarly obtaining the average distance of each segmentation result, taking the segmentation result corresponding to the maximum value of the average distance as the best segmentation result, taking the last initial window in the best segmentation result in chronological order as the target window, and the end of the target window as the time to be tested.

[0008] Preferably, the first distinguishing distance includes: taking data segments other than the target segment as reference segments; and for any principal component, taking the DTW distance between the projected value vector of the target segment and the projected value vector of the reference segment as the first distinguishing distance between the target segment and the reference segment in any principal component.

[0009] Preferably, the second discrimination distance includes: taking data segments other than the target segment as reference segments; for any principal component, calculating the absolute value of the difference between the loading of the target segment in any dimension and the loading of the reference segment in any dimension, and calculating the first product of the absolute value of the difference and the recognition saliency in any dimension, traversing to obtain the first product of the target segment and the reference segment in each dimension, and summing all the first products as the second discrimination distance between the target segment and the reference segment in any principal component.

[0010] Preferably, calculating the recognition contribution of the target segment includes: taking data segments other than the target segment as reference segments; for any principal component, calculating the second product of the first discrimination distance and the second discrimination distance, traversing to obtain the second product of the target segment and the reference segment in each principal component, accumulating the value of all second products as the second accumulated value, similarly obtaining the second accumulated value of the target segment and each reference segment, and summing all the second accumulated values ​​as the recognition contribution of the target segment.

[0011] Preferably, calculating the recognition contribution of the target segment includes: for any principal component, using the DTW distance between the projected value vector of the target segment and the projected value vector of the target window as the first distinguishing distance between the target segment and the target window in any principal component; for any principal component, calculating the absolute value of the difference between the loading of the target segment in any dimension and the loading of the target window in any dimension, and calculating the third product of the absolute value of the difference and the recognition saliency in any dimension, iterating through the third products of the target segment and the target window in each dimension, and summing all the third products as the second distinguishing distance between the target segment and the target window in any principal component; calculating the first sum of the first distinguishing distance and the second distinguishing distance; calculating the sum of the products of the loading vector of the target segment in any principal component and the recognition saliency in each dimension, iterating through the sum of the products of each principal component, and using the ratio of the sum of the products of any principal component to the sum of the sums of the products of all principal components as the principal component weight; calculating the fourth product of the principal component weight of any principal component and the first sum, and using the sum of the fourth products of all principal components in the target segment as the recognition contribution.

[0012] Preferably, calculating the fault degree of the refrigerator to be tested includes: for any principal component, calculating the second distinguishing distance between the refrigerator to be tested and any normal refrigerator in the target segment; taking the average of the second distinguishing distances between the refrigerator to be tested and any normal refrigerator in all data segments as the first average; similarly obtaining the second average of the refrigerator to be tested and any abnormal refrigerator in all data segments; calculating the difference between the first average and the second average, and calculating the exponent value of the difference; iterating through each principal component to obtain the exponent value; calculating the second sum of all exponent values; normalizing the identification contribution of the target segment to obtain a normalized value; calculating the fifth product of the normalized value and the second sum; iterating through each data segment to obtain the fifth product; and taking the sum of all fifth products as the fault degree of the refrigerator to be tested.

[0013] In a second aspect, a fault detection system for a refrigeration machine includes: a processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement any of the fault detection methods for a refrigeration machine described in the previous article.

[0014] The present invention has the following effects:

[0015] This invention, by introducing dynamic collaborative analysis of multi-dimensional parameters and principal component-based dimensionality reduction technology, can effectively capture early weak abnormal signals of refrigeration machine failures and accurately quantify the occurrence and development process of the failure.

[0016] By refining the evaluation of the contribution of each data segment to fault detection, the problem of fault signals being masked by periodic fluctuations, as seen in traditional methods, is avoided. By combining dynamic time warping distance and principal component load structure differences, not only can key data segments significantly different from normal fluctuations be identified, but the sensitivity to progressive faults is also enhanced while ensuring the full preservation of the dynamic characteristics of the time series, making fault detection more timely and accurate. Attached Figure Description

[0017] Figure 1 This is a flowchart of a fault detection method for a refrigeration machine according to an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0019] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0020] Reference Figure 1 A fault detection method for a refrigeration machine includes steps S1-S3, as detailed below:

[0021] S1: Obtain the multi-dimensional parameter set of any refrigeration machine in history and calculate the recognition saliency of any dimension.

[0022] In one embodiment, for several refrigerators in history, a multi-dimensional parameter set of any refrigerator in history is obtained. The multi-dimensional parameters include high-pressure side pressure and low-pressure side pressure (reflecting the refrigerant circulation state), suction temperature and discharge temperature (characterizing the thermodynamic efficiency of the compressor), condenser inlet air temperature and evaporator return air temperature (reflecting the heat exchange environment and load matching), and vibration amplitude of the compressor body (used to monitor mechanical wear, bearing failure or resonance risk).

[0023] It should be noted that in refrigeration unit fault detection, common faults such as refrigerant leakage, ice blockage, and compressor wear are often not isolated manifestations of a single parameter anomaly, but rather are reflected through the coordinated changes of multiple operating parameters such as pressure, temperature, and vibration. Existing technologies often rely on the static coupling relationship between pairs of parameters, performing real-time analysis within a fixed time window. This approach ignores the non-uniform multi-parameter coupling chain characteristics and significant time-series dependence inherent in the fault evolution of refrigeration systems. For example, condenser fouling will sequentially lead to an increase in condensing temperature → an increase in condensing pressure → an increase in the high-low pressure ratio → a rise in discharge temperature → an increase in compressor power consumption, forming a dynamic coupling chain involving five dimensions of parameters with a clear causal sequence. In contrast, refrigerant insufficiency may only manifest as a simple binary relationship between a decrease in suction pressure and an increase in suction temperature. Because existing technologies struggle to distinguish the complexity and evolution path of coupling chains under different fault modes, they lack sensitivity to multi-parameter coordinated anomalies, easily overlooking early fault characteristics, resulting in low fault detection sensitivity and delayed diagnosis.

[0024] PCA (Principal Component Analysis) is used to reduce the dimensionality of the multi-dimensional parameter set of any refrigeration machine in history to obtain the eigenvalues ​​of each principal component and the loadings of each dimension. The loadings of all dimensions in the same principal component are then constructed into a loading vector.

[0025] Using any dimension as the target dimension, the product of the absolute value of the target dimension loading and the eigenvalue of any principal component is calculated as the dimension weight of the target dimension. The sum of the dimension weights of all principal components in the target dimension is taken as the first salient distance of the target dimension.

[0026] The load vectors corresponding to the principal components whose target dimension loads are not equal to 0 are taken as valid vectors. The refrigerators in history are divided into normal refrigerators and abnormal refrigerators. The valid vector of any abnormal refrigerator is taken as the abnormal valid vector, and the valid vector of any normal refrigerator is taken as the normal valid vector. The absolute similarity value between any abnormal valid vector and any normal valid vector is calculated. The sum of the absolute similarity values ​​between any abnormal valid vector and each normal valid vector is taken as the first accumulated value. The first accumulated value of each abnormal valid vector is obtained by traversing. The sum of all the first accumulated values ​​is taken as the second significant distance of the target dimension.

[0027] Calculate the absolute difference between the first saliency distance of any abnormal refrigerator and the first saliency distance of any normal refrigerator, and calculate the ratio of the absolute difference to the second saliency distance. Iterate through the ratios of any abnormal refrigerator to each normal refrigerator, and sum all the ratios as the recognition saliency of the target dimension.

[0028] It should be explained that the difference in the first significant distance is used to quantify the difference in the relative contribution of the abnormal refrigerator and the normal refrigerator in the target dimension to the overall multi-dimensional coupling structure: the larger the difference, the more significant the influence of the parameter of the target dimension on the coupling relationship before and after the failure occurs; while the second significant distance represents the structural difference in the load vector corresponding to the parameter of the dimension in the fault state and the normal state in the multi-dimensional coupling space, assuming that the target dimension does indeed carry effective information (i.e. has a non-zero load). It reflects not the change magnitude of a single dimension, but the fundamental shift in the interaction mode between dimensions.

[0029] S2: Collect a multi-dimensional parameter set of the refrigerator to be tested, construct a target window for the refrigerator to be tested, with the end of the target window being the time to be tested. Divide the target window into segments to obtain several data segments. Take any data segment as the target segment, perform PCA dimensionality reduction on the target segment to obtain principal components, projection value vectors, and loading vectors. One principal component corresponds to one projection value vector and one loading vector. Calculate the first discrimination distance based on the projection value vector, and calculate the second discrimination distance based on the loading vector and the recognition saliency. Calculate the recognition contribution of the target segment based on the first discrimination distance and the second discrimination distance.

[0030] In one embodiment, the multi-dimensional parameter set of the refrigerator to be tested is randomly divided into several initial windows of unequal length. For a segmentation result, the DTW (Dynamic Time Warping) distance between any two window parameter sets is calculated, and the average distance of all DTW distances is calculated. Similarly, the average distance of each segmentation result is obtained. The segmentation result corresponding to the maximum value of the average distance is taken as the best segmentation result. The last initial window in the best segmentation result in chronological order is taken as the target window, and the end of the target window is the time to be detected.

[0031] It should be explained that the segmentation result corresponding to the maximum mean represents the window division method that best matches the variation characteristics of the parameter set. This segmentation method can effectively select the cut-off point that best matches the time series variation pattern from multiple possible segmentation results, and use this as the basis to determine the detection time, thereby providing more accurate refrigeration unit status monitoring. This method not only optimizes the parameter window selection process but also ensures the accuracy of the detection time.

[0032] The target window is segmented into several data segments with a fixed length. For example, the fixed length is the reciprocal of the fundamental frequency of the refrigerator. Any data segment is used as the target segment, and PCA dimensionality reduction is performed on the target segment to obtain principal components, projected value vectors, and load vectors. Each principal component corresponds to one projected value vector and one load vector.

[0033] Data segments other than the target segment are used as reference segments; for any principal component, the DTW distance between the projected value vector of the target segment and the projected value vector of the reference segment is used as the first distinguishing distance between the target segment and the reference segment in any principal component.

[0034] For any principal component, calculate the absolute value of the difference between the loading of the target segment in any dimension and the loading of the reference segment in any dimension, and calculate the first product of the absolute value of the difference and the recognition saliency in any dimension. Iterate through the first products of the target segment and the reference segment in each dimension, and use the sum of all first products as the second discriminant distance between the target segment and the reference segment in any principal component.

[0035] Calculate the second product of the first and second discrimination distances, iterate through the target segment and the reference segment to obtain the second product of each principal component, and use the sum of all the second products as the second accumulated value. Similarly, obtain the second accumulated value of the target segment and each reference segment, and use the sum of all the second accumulated values ​​as the recognition contribution of the target segment.

[0036] In another embodiment, for any principal component, the DTW distance between the projected value vector of the target segment and the projected value vector of the target window is used as the first distinguishing distance between the target segment and the target window in any principal component.

[0037] For any principal component, calculate the absolute value of the difference between the loading of the target segment in any dimension and the loading of the target window in any dimension, and calculate the third product of the absolute value of the difference and the recognition saliency in any dimension. Iterate through the target segment and the target window to obtain the third product in each dimension, and sum all the third products as the second discriminant distance between the target segment and the target window in any principal component.

[0038] Calculate the first sum of the first and second discrimination distances; calculate the sum of the products of the target segment's loading vector on any principal component and the recognition saliency of each dimension, iterate through each principal component to obtain the sum of the products, and use the ratio of the sum of the products of any principal component to the cumulative sum of the products of all principal components as the principal component weight.

[0039] Calculate the fourth product of the principal component weight of any principal component and the first sum, and use the sum of the fourth products of all principal components in the target segment as the identification contribution.

[0040] This embodiment effectively improves the accuracy and robustness of matching and identification between target segments and target windows by integrating temporal morphological similarity and principal component load structure differences. On the one hand, it uses DTW distance to describe the dynamic temporal alignment similarity of the two on the principal component projection trajectory, fully preserving the temporal evolution characteristics of the original data. On the other hand, it introduces a second distinguishing distance weighted by dimensional load differences and identification saliency, reflecting the sensitivity differences of the two on key variables from the feature space structure level, making the distinction more physically interpretable. On this basis, the principal component weights guided by identification saliency adaptively adjust the influence of each principal component on the identification contribution, highlighting the principal components that play a dominant role in system state discrimination. Finally, the identification contribution formed by combining temporal and structural information can more accurately and stably quantify the matching degree between target segments and target windows.

[0041] S3: Calculate the degree of fault of the refrigerator to be detected based on the identification contribution and the second distinguishing distance, and complete the fault detection.

[0042] It should be noted that in the fault detection of refrigeration machines, traditional refrigeration machines often achieve energy efficiency optimization through dynamic frequency adjustment, resulting in periodic fluctuations in operating parameters. Early or mid-stage faults often manifest as slow, gradual performance degradation, and their fault characteristics are easily masked by these normal fluctuations in the early stages. Existing technologies typically use a fixed window to process data uniformly, ignoring the differences in the contribution of data from different time periods within the window to fault identification. Especially when the fault is still in its development stage, key weak abnormal signals are easily diluted by a large amount of normal or similarly fluctuating data within the window, leading to smoothed fault characteristics, underestimated fault severity, and consequently, detection delays or even missed detections. To address this issue, this invention proposes a fault severity quantification method based on a combination of identification contribution and a second distinguishing distance. By finely evaluating the actual contribution of each data segment within a fixed window to fault discrimination, it highlights sensitive segments that reflect the true degradation trend. Simultaneously, by combining principal component load structure differences and dimensional identification saliency, it effectively distinguishes between normal fluctuations and early anomalies, thereby significantly improving the sensitivity and detection timeliness of mid-stage gradual faults while preserving the periodic dynamic characteristics.

[0043] In one embodiment, for any principal component, the second distinguishing distance between the refrigerator under test and any normal refrigerator in the target segment is calculated. The average of the second distinguishing distances between the refrigerator under test and any normal refrigerator in all data segments is taken as the first mean. Similarly, the second mean between the refrigerator under test and any abnormal refrigerator in all data segments is obtained. The difference between the first mean and the second mean is calculated, and the exponent of the difference is calculated. The exponent of each principal component is obtained through iteration, and the second sum of all exponents is calculated. The identification contribution of the target segment is normalized to obtain a normalized value. The fifth product of the normalized value and the second sum is calculated. The fifth product of each data segment is obtained through iteration, and the sum of all fifth products is taken as the fault degree of the refrigerator under test.

[0044] The degree of contribution is measured by the relative importance of each data segment within a fixed window for fault detection. Its core is to distinguish the actual role of data in reflecting fault characteristics in different time periods: during the operation of the refrigeration unit, some key data segments may contain early abnormal fluctuations, transient deviations or sudden changes in sensitive variables, while other time periods mainly show normal periodic changes or steady-state operation. If the traditional equal weighting method is used, these abnormal segments with high discrimination value will be diluted by a large amount of conventional data, resulting in the overall fault signal being weakened.

[0045] This invention constructs an identification contribution model that comprehensively considers the dynamic projection differences of each data segment in the principal component space and the load structure sensitivity, assigning higher contribution values ​​to data segments that significantly deviate from the normal pattern and are located in the direction of high-weight principal components, thereby avoiding fault assessment being dominated by long-term average trends or strong periodicity. On this basis, the identification contribution of each data segment is weighted and fused, which can more realistically restore the stage characteristics of fault development, highlight local abnormal responses, and effectively improve the perception sensitivity of the early and middle stages of progressive faults.

[0046] When the fault level of the refrigerator under test exceeds the preset fault threshold, an alarm signal is generated and sent to complete the fault detection.

[0047] The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement a fault detection method for a refrigeration machine according to the first aspect of the present invention.

[0048] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0049] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A fault detection method for a refrigeration machine, characterized in that, include: Obtain the multi-dimensional parameter set of any refrigeration machine in history, and calculate the recognition saliency of any dimension; A multi-dimensional parameter set of the refrigerator under test is collected to construct a target window for the refrigerator under test. The end of the target window is the time to be tested. The target window is segmented to obtain several data segments. Any data segment is used as the target segment. PCA dimensionality reduction is performed on the target segment to obtain principal components, projection value vectors, and loading vectors. One principal component corresponds to one projection value vector and one loading vector. The first discrimination distance is calculated based on the projection value vector. The second discrimination distance is calculated based on the loading vector and the recognition saliency. The recognition contribution of the target segment is calculated based on the first discrimination distance and the second discrimination distance. The degree of fault of the refrigerator under test is calculated based on the identification contribution and the second distinguishing distance, and the fault detection is completed. Calculating the saliency of any dimension includes: PCA dimensionality reduction is performed on the multi-dimensional parameter set of any refrigerator in history to obtain the eigenvalues ​​of each principal component and the loads of each dimension. The loads of all dimensions in the same principal component are then constructed into a load vector. Take any dimension as the target dimension, calculate the product of the absolute value of the target dimension loading and the eigenvalue of any principal component as the dimension weight of the target dimension, and take the sum of the dimension weights of all principal components in the target dimension as the first salient distance of the target dimension. The load vectors corresponding to the principal components whose target dimension loads are not equal to 0 are taken as valid vectors. The refrigerators in history are divided into normal refrigerators and abnormal refrigerators. The valid vector of any abnormal refrigerator is taken as an abnormal valid vector, and the valid vector of any normal refrigerator is taken as a normal valid vector. The absolute value of the similarity between any abnormal valid vector and any normal valid vector is calculated. The sum of the absolute values ​​of the similarity between any abnormal valid vector and each normal valid vector is taken as the first accumulated value. The first accumulated value of each abnormal valid vector is obtained by traversing. The sum of all the first accumulated values ​​is taken as the second significant distance of the target dimension. Calculate the absolute difference between the first saliency distance of any abnormal refrigerator and the first saliency distance of any normal refrigerator, and calculate the ratio of the absolute difference to the second saliency distance. Iterate through the ratios of any abnormal refrigerator to each normal refrigerator, and sum all the ratios as the recognition saliency of the target dimension. The target window for constructing the refrigerator to be tested includes: The multi-dimensional parameter set of the refrigerator to be tested is randomly divided into several initial windows of unequal length; For a segmentation result, calculate the DTW distance between any two window parameter sets, and calculate the average distance of all DTW distances. Similarly, obtain the average distance of each segmentation result, take the segmentation result corresponding to the maximum value of the average distance as the best segmentation result, take the last initial window in the best segmentation result in chronological order as the target window, and take the end of the target window as the time to be detected. The first distinguishing distance includes: Use data segments other than the target segment as reference segments; For any principal component, the DTW distance between the projected value vector of the target segment and the projected value vector of the reference segment is taken as the first distinguishing distance between the target segment and the reference segment in any principal component. The second distinguishing distance includes: Use data segments other than the target segment as reference segments; For any principal component, calculate the absolute value of the difference between the loading of the target segment in any dimension and the loading of the reference segment in any dimension, and calculate the first product of the absolute value of the difference and the recognition saliency in any dimension. Iterate through the first products of the target segment and the reference segment in each dimension, and use the sum of all first products as the second discriminant distance between the target segment and the reference segment in any principal component.

2. The fault detection method for a refrigeration machine according to claim 1, characterized in that, Calculating the recognition contribution of the target segment includes: Use data segments other than the target segment as reference segments; For any principal component, calculate the second product of the first and second distinguishing distances. Iterate through the target segment and the reference segment to obtain the second product of each principal component. The sum of all the second products is taken as the second accumulated value. Similarly, the second accumulated value of the target segment and each reference segment is obtained. The sum of all the second accumulated values ​​is taken as the recognition contribution of the target segment.

3. The fault detection method for a refrigeration machine according to claim 1, characterized in that, Calculating the recognition contribution of the target segment includes: For any principal component, the DTW distance between the projected value vector of the target segment and the projected value vector of the target window is taken as the first distinguishing distance between the target segment and the target window in any principal component. For any principal component, calculate the absolute value of the difference between the loading of the target segment in any dimension and the loading of the target window in any dimension, and calculate the third product of the absolute value of the difference and the recognition saliency in any dimension. Iterate through the target segment and the target window to obtain the third product in each dimension, and use the sum of all the third products as the second discriminant distance between the target segment and the target window in any principal component. Calculate the first sum of the first and second discrimination distances; Calculate the sum of the products of the target segment's loading vector at any principal component and the recognition saliency of each dimension. Iterate through each principal component to obtain the sum of the products. Use the ratio of the sum of the products of any principal component to the cumulative sum of the products of all principal components as the principal component weight. Calculate the fourth product of the principal component weight of any principal component and the first sum, and use the sum of the fourth products of all principal components in the target segment as the identification contribution.

4. The fault detection method for a refrigeration machine according to claim 1, characterized in that, Calculating the fault severity of the refrigerator under test includes: For any principal component, calculate the second distinguishing distance between the refrigerator under test and any normal refrigerator in the target segment. Take the average of the second distinguishing distances between the refrigerator under test and any normal refrigerator in all data segments as the first mean. Similarly, obtain the second mean between the refrigerator under test and any abnormal refrigerator in all data segments. Calculate the difference between the first mean and the second mean, and calculate the exponent of the difference. Iterate through each principal component to obtain the exponent value, and calculate the second sum of all exponent values. The recognition contribution of the target segment is normalized to obtain a normalized value, and the fifth product of the normalized value and the second sum is calculated. Iterate through each data segment to obtain the fifth product, and sum all the fifth products as the degree of failure of the refrigerator to be tested.

5. A fault detection system for a refrigeration unit, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a fault detection method for a refrigeration machine according to any one of claims 1-4.

Images