Refrigeration equipment fault prediction and diagnosis method, device and system

Abstract

The invention discloses a refrigeration equipment fault prediction and diagnosis method, device and system, and the method comprises the steps: collecting a vibration signal and a working condition signal of refrigeration equipment, carrying out the preprocessing of the original data of the vibration signal and the working condition signal, and obtaining the vibration characteristic data and the working condition data; aligning the vibration characteristic data with the working condition data, dividing working conditions according to the working condition data, and classifying the vibration characteristic data under each working condition to obtain a plurality of groups of fault diagnosis data sets; on the basis of each working condition, historical healthy vibration data of the refrigeration equipment are obtained, a fault diagnosis model is constructed and trained, abnormal diagnosis is conducted on the vibration characteristic data through the trained fault diagnosis model, and abnormal fault points are obtained; and fault positioning analysis is carried out by adopting a spectrum analysis method, fault positioning information corresponding to each working condition is obtained, and a maintenance work order of the refrigeration equipment is generated and displayed. According to the method, accurate fault pre-judgment diagnosis can be carried out on the refrigeration equipment, and maintenance is convenient.

Description

Technical Field

[0001] The present application relates to the technical field of refrigeration equipment maintenance, and in particular to a method, device and system for predicting and diagnosing refrigeration equipment faults. Background Art

[0002] Refrigeration equipment includes refrigeration compressors, ice machines and other machinery and equipment. Refrigeration compressors are an indispensable power source in industrial production. During long-term operation, refrigeration compressors may cause equipment performance degradation, reduced efficiency, sudden failures and other problems due to normal wear, overload, improper maintenance or operational errors. In severe cases, production interruptions may affect corporate benefits. Therefore, the implementation of effective equipment vibration monitoring and fault warning has become the key to ensuring the normal operation of equipment.

[0003] In the traditional refrigeration system fault prediction scheme, the vibration signal of the refrigeration compressor is collected, and the vibration characteristics are used for abnormal detection, fault diagnosis, and fault location through a general algorithm model. However, the training of general algorithm models often relies on normal healthy samples and abnormal fault samples. In the actual use of refrigeration equipment, abnormal fault samples are difficult to obtain. Therefore, the use of general algorithm models to perform abnormal diagnosis of vibration characteristics will result in low detection accuracy and inaccurate fault prediction due to the lack of abnormal fault samples, and cannot play a positive role in actual early warning maintenance. Summary of the invention

[0004] The embodiments of the present application provide a method, device and system for predicting and diagnosing faults of refrigeration equipment, which do not rely on abnormal fault samples to train the algorithm model, and can perform relatively accurate predictive fault diagnosis on the refrigeration equipment, which is beneficial to the inspection and maintenance of the refrigeration equipment.

[0005] In a first aspect, an embodiment of the present application provides a method for predicting and diagnosing faults of refrigeration equipment, the method comprising the following steps: collecting vibration signals and operating condition signals of key components of the refrigeration equipment under various operating conditions, preprocessing original vibration characteristic data in the vibration signal and original operating condition data in the operating condition signal, respectively, to obtain vibration characteristic data and operating condition data;

[0006] Aligning the vibration characteristic data with the operating condition data, dividing the operating condition according to the operating condition data, and classifying the vibration characteristic data based on each operating condition to obtain multiple groups of fault diagnosis data sets;

[0007] Based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, and a fault diagnosis model is constructed and trained using the historical healthy vibration data and the vibration characteristic data to be tested. The fault diagnosis model is used to perform abnormal diagnosis on the vibration characteristic data of the corresponding working condition to obtain abnormal fault points, including: based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, and the historical healthy vibration data is used to train an initial OCSVM model to determine an optimal dictionary matrix and a sparse representation matrix, and the vibration characteristic data to be tested of the corresponding working condition is input into the initial OCSVM model, and the sparse representation matrix is ​​updated based on the vibration characteristic data to be tested, wherein the multi-dimensional matrix of the initial OCSVM model is updated. The kernel mixed kernel function is obtained by fusing the Gaussian kernel function and the linear kernel function; based on the multi-core mixed kernel function, the initial OCSVM model is reconstructed through the optimal dictionary matrix and the updated sparse representation matrix to obtain the fault diagnosis model; based on the decision boundary of the fault diagnosis model, the abnormality detection score of each vibration feature data is calculated, and the initial abnormality detection threshold is adjusted according to the abnormality detection score to obtain the dynamic threshold; the vibration feature data with the abnormality detection score greater than the dynamic threshold is taken as the abnormal fault point; for the abnormal fault point under each working condition, the spectrum analysis method is used to perform fault location analysis, to obtain the fault location information corresponding to each working condition, and to generate and display the maintenance work order of the refrigeration equipment.

[0008] In a second aspect, an embodiment of the present application provides a refrigeration equipment fault prediction and diagnosis device, comprising:

[0009] The signal acquisition and processing module is used to collect the vibration signals and working condition signals of each key component of the refrigeration equipment under various working conditions, and pre-process the original vibration characteristic data in the vibration signal and the original working condition data in the working condition signal to obtain the vibration characteristic data and working condition data;

[0010] A working condition classification module, used for aligning the vibration characteristic data with the working condition data, classifying the working conditions according to the working condition data, and classifying the vibration characteristic data under each working condition to obtain multiple groups of fault diagnosis data sets;

[0011] An abnormality detection module is used to obtain historical healthy vibration data of the refrigeration equipment based on each working condition, use the historical healthy vibration data and the vibration characteristic data to be tested to build and train a fault diagnosis model, and perform abnormal diagnosis on the vibration characteristic data of the corresponding working condition through the trained fault diagnosis model to obtain an abnormal fault point;

[0012] The fault diagnosis module is used to perform fault location analysis on the abnormal fault point under each working condition by using a spectrum analysis method, obtain fault location information corresponding to each working condition, generate a maintenance work order for the refrigeration equipment and display it.

[0013] Wherein, the anomaly detection module includes:

[0014] A data acquisition unit is used to acquire historical healthy vibration data of the refrigeration equipment based on each working condition, use the historical healthy vibration data to train an initial OCSVM model, determine an optimal dictionary matrix and a sparse representation matrix, input the vibration characteristic data to be measured corresponding to the working condition into the initial OCSVM model, and update the sparse representation matrix based on the vibration characteristic data to be measured, wherein the multi-core hybrid kernel function of the initial OCSVM model is obtained by fusion of a Gaussian kernel function and a linear kernel function;

[0015] A model reconstruction unit, used to reconstruct the initial OCSVM model through the optimal dictionary matrix and the updated sparse representation matrix based on a multi-core hybrid kernel function to obtain a fault diagnosis model;

[0016] A threshold adjustment unit, configured to calculate an abnormality detection score for each of the vibration characteristic data based on a decision boundary of the fault diagnosis model, and to adjust an initial abnormality detection threshold according to the abnormality detection score to obtain a dynamic threshold;

[0017] A fault point detection unit, used to take the vibration characteristic data whose abnormal detection score is greater than the dynamic threshold as an abnormal fault point;

[0018] On the third aspect, the embodiments of the present application also provide a refrigeration equipment fault prediction and diagnosis system, including: a vibration sensor for collecting vibration signals of the refrigeration equipment under various operating conditions; an operating condition sensor for collecting operating condition signals of the refrigeration compressor under various operating conditions; and a server for performing predictive fault diagnosis on the refrigeration equipment based on the vibration signal and the operating condition signal, using the refrigeration equipment fault prediction and diagnosis method described in the above embodiments.

[0019] In an embodiment of the present application, vibration detection and operating status detection are performed on the refrigeration compressor of the refrigeration equipment under complex working conditions to obtain vibration signals and working condition signals, and the working conditions are divided based on the preprocessed vibration characteristic data and working condition data, and fault diagnosis is performed jointly, thereby eliminating interference between multiple vibration characteristics between different working conditions, reducing misjudgment of normal vibration characteristics between different working conditions, and improving the accuracy of fault diagnosis; and, based on the historical healthy vibration data of each working condition, a fault diagnosis model is constructed and trained, and the vibration characteristic data under each working condition is diagnosed abnormally through the trained fault diagnosis model. Only positive sample training is used in model training, which can avoid the problem of low accuracy due to the lack of abnormal fault samples, and is more conducive to the inspection and maintenance of refrigeration equipment.

[0020] Furthermore, the fault diagnosis model is obtained by improving the OCSVM model, which can continue the advantages of the OCSVM model to improve the feature expression ability, enhance the model robustness, and better adapt to the data distribution of different working conditions; at the same time, the fault diagnosis model based on multi-core function fusion can express features more comprehensively, provide more comprehensive and diversified feature expression capabilities, and can adapt to vibration and working condition data distributed in different data types, reduce the fault diagnosis model's dependence on a single function, and make the model more robust.

[0021] Furthermore, when diagnosing abnormal fault points, fixed threshold comparison is abandoned and dynamic threshold comparison is adopted. The dynamic threshold can be flexibly adjusted, which enables the model to better adapt to the real-time changes of the vibration data and operating data of the refrigeration compressor, improves the real-time performance and accuracy of the algorithm, and reduces the false alarm rate.

[0022] Furthermore, the healthy vibration data of the refrigeration equipment can be updated through the feedback information of the maintenance work order, and the fault diagnosis model can be iteratively trained through the updated healthy vibration data, and the abnormal detection threshold can be continuously corrected, so that the fault diagnosis model can predict the fault point more accurately as the use time goes by. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of a scenario of a method for predicting and diagnosing refrigeration equipment faults according to an embodiment of the present application.

[0025] Figure 2 It is a flow chart of the refrigeration equipment fault prediction and diagnosis method according to an embodiment of the present application.

[0026] Figure 3 It is a schematic diagram of the installation of each vibration sensor in the embodiment of the present application.

[0027] Figure 4 It is a flowchart of step S3 of the method for predicting and diagnosing refrigeration equipment faults in an embodiment of the present application.

[0028] Figure 5 This is a schematic diagram of abnormal fault point diagnosis in an embodiment of the present application.

[0029] Figure 6 It is a structural schematic diagram of a refrigeration equipment fault prediction and diagnosis device according to an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0032] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0033] The refrigeration equipment fault prediction and diagnosis method of the embodiment of the present application is applied to the refrigeration equipment fault prediction and diagnosis system, such as Figure 1 As shown, the refrigeration equipment fault prediction and diagnosis system includes a vibration sensor 101, a working condition sensor 102 and a server 103, wherein the vibration sensor 101 is installed on a key component of the refrigeration equipment that is prone to failure, and is used to collect vibration signals of each key component of the refrigeration equipment under various working conditions or variable working conditions; the working condition sensor 102 is installed on key components such as control channels or control valves on the refrigeration equipment, and is used to collect working condition signals of each key component of the refrigeration equipment under various working conditions or variable working conditions; the server 103, which can be a cloud server or a physical computer server, is used to perform predictive fault diagnosis on the refrigeration equipment using a refrigeration equipment fault prediction and diagnosis method. Both the vibration sensor 101 and the working condition sensor 102 are connected to the edge gateway 104 through wires to achieve communication, and the collected vibration signals and working condition signals are uploaded to the server 103 through the edge gateway 104, and then the server 103 performs data analysis and processing to achieve predictive fault diagnosis of the refrigeration equipment.

[0034] Refrigeration equipment includes refrigeration compressors, ice machines and other machinery and equipment. The refrigeration equipment fault prediction and diagnosis method of the present application is applicable to refrigeration compressors, ice machines and other refrigeration equipment. The following is a detailed description using a refrigeration compressor as an example. The refrigeration compressor is the core component in the refrigeration system. Its main function is to compress the low-pressure, low-temperature refrigerant gas from the evaporator into a high-pressure, high-temperature gas, and send it to the condenser for cooling. This process is a key step in the refrigeration cycle (such as a vapor compression refrigeration cycle). It uses compressor refrigerant vapor to achieve heat transfer from a low-temperature environment to a high-temperature environment. The normal operation of the refrigeration compressor is crucial to the performance of the entire refrigeration system. When a compressor fails, it is necessary to perform a fault diagnosis in a timely manner so that corresponding repair measures can be taken. If there is no compressor on the refrigeration equipment and other refrigeration methods are used, the fault prediction and diagnosis method is still applicable.

[0035] like Figure 2 As shown, the refrigeration equipment fault prediction and diagnosis method of this embodiment includes the following steps:

[0036] S1: Collect vibration signals and working condition signals of key components of refrigeration equipment under various working conditions, pre-process original vibration characteristic data in the vibration signal and original working condition data in the working condition signal, and obtain vibration characteristic data and working condition data.

[0037] In this embodiment, the refrigeration equipment is a refrigeration compressor, and the working conditions of the refrigeration compressor may include pressure, flow, and temperature. The pressure of the refrigeration compressor is one of the important indicators for measuring the operating status of the refrigeration system. Different refrigeration systems and different refrigerants will have different pressure ranges, which can be divided into different working conditions such as primary pressure, secondary pressure, and tertiary pressure according to the pressure. If pressure is used as the monitoring working condition, the working condition data may include pressure values ​​such as the exhaust system pressure and the bearing oil pressure. The flow of the refrigeration compressor refers to the amount of refrigerant volume or mass that the compressor can handle per unit time, which can directly affect the efficiency and performance of the refrigeration system. It can be divided into different working conditions such as primary flow, secondary flow, and tertiary flow according to the flow. In addition, if temperature is used as the monitoring working condition, the working condition data may also include temperature values ​​such as the secondary intake temperature, the primary exhaust temperature, the secondary exhaust temperature, and the bearing oil temperature.

[0038] The vibration signals of each key component of the refrigeration compressor under various working conditions are collected by vibration sensors. The vibration signals are used to analyze the vibration characteristics of the refrigeration equipment. The working condition signals are collected by working condition sensors. The working condition signals are used to divide the working conditions so that the characteristic values ​​of the vibration signals can be classified according to the working conditions. The refrigeration equipment of the refrigeration system, especially the refrigeration compressor, continues to operate under variable working conditions, and usually there is no clear change pattern. During the collection process of vibration signals and working condition signals, although they are collected by corresponding sensors, it is still impossible to clearly distinguish which working condition the vibration signal belongs to. In other words, the vibration signals and working condition signals are collected in real time at each moment. The collected signals are mixed with normal or abnormal vibration signals, normal or abnormal working condition signals, various fault noise signals, and different vibration signals and working condition signals between multiple working conditions, thus forming an original data set with a large amount of data and high complexity.

[0039] Since there is a certain amount of noise data in the collected vibration signals and operating condition signals, preprocessing is performed before fault point diagnosis is performed through the algorithm model. That is, the interference noise needs to be removed as much as possible to improve the accuracy of the algorithm model in predicting and diagnosing refrigeration equipment faults.

[0040] S2: Align the vibration characteristic data with the working condition data, divide the working condition according to the working condition data, classify the vibration characteristic data under each working condition, and obtain multiple groups of fault diagnosis data sets.

[0041] Since the vibration signal and the operating condition signal are collected at different frequencies, there will be a certain time difference between the vibration signal and the operating condition signal in the original data. Therefore, it is necessary to align the vibration feature data and the operating condition data, and add timestamps to the aligned vibration feature data and operating condition data, so that the vibration feature data can be more accurately classified based on the operating condition and the fault diagnosis data set can be divided.

[0042] S3: Based on each working condition, the historical healthy vibration data of the refrigeration equipment is obtained, and a fault diagnosis model is constructed and trained using the historical healthy vibration data and the vibration characteristic data to be tested. The vibration characteristic data of the corresponding working condition is subjected to abnormal diagnosis through the trained fault diagnosis model to obtain the abnormal fault point.

[0043] The fault diagnosis model is obtained by improving the OCSVM model, and the kernel function of the fault diagnosis model is a multi-core hybrid kernel function obtained by fusing a Gaussian kernel function and a linear kernel function.

[0044] In this embodiment, step S3 may include the following steps:

[0045] S31: Based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, the historical healthy vibration data is used to train the initial OCSVM model, the optimal dictionary matrix and the sparse representation matrix are determined, the vibration characteristic data to be measured corresponding to the working condition is input into the initial OCSVM model, and the sparse representation matrix is ​​updated based on the vibration characteristic data to be measured. The multi-core hybrid kernel function of the initial OCSVM model is obtained by fusion of the Gaussian kernel function and the linear kernel function.

[0046] S32: Based on the multi-core hybrid kernel function, reconstruct the initial OCSVM model through the optimal dictionary matrix and the updated sparse representation matrix to obtain a fault diagnosis model.

[0047] S33: Based on the decision boundary of the fault diagnosis model, the abnormality detection score of each vibration feature data is calculated, and the initial abnormality detection threshold is adjusted according to the abnormality detection score to obtain a dynamic threshold.

[0048] S34: The vibration characteristic data whose abnormal detection score is greater than the dynamic threshold is taken as an abnormal fault point.

[0049] Generally speaking, it is difficult to collect abnormal fault data of refrigeration compressors. Therefore, this embodiment uses historical healthy vibration data to train the algorithm model, thereby avoiding the pain points and difficulties of abnormal vibration data collection. The abnormal fault point is the characteristic value of the vibration signal collected when a key component of the refrigeration compressor fails. This characteristic value will be different from the normal characteristic value. The larger the difference, the larger and more obvious the fault is, and the smaller the difference, the smaller and more hidden the fault is. However, if the equipment fails, no matter how big the fault is, it will bring safety hazards. Therefore, this abnormal fault point needs to be accurately identified.

[0050] This embodiment detects abnormal data by jointly analyzing the working condition data and the vibration data, which can reduce the impact of the working condition on the accuracy of diagnosis. Using a dynamic threshold for abnormal point diagnosis can improve the adaptability of the model.

[0051] S4: For the abnormal fault point under each working condition, the spectrum analysis method is used to perform fault location analysis, obtain the fault location information corresponding to each working condition, generate and display the maintenance work order of the refrigeration equipment.

[0052] In the embodiments of the present application, establishing different fault diagnosis models after dividing the vibration feature data by the working condition data has the following beneficial effects: 1. Enhancing the specificity and accuracy of the fault diagnosis model: Different working conditions may cause the vibration characteristics of the refrigeration equipment to be very different. Establishing a fault diagnosis model separately for each working condition allows the model to focus more on the data features under specific conditions, thereby improving the accuracy of anomaly detection. 2. Reducing model complexity: By dividing the vibration feature data into working conditions, each fault diagnosis model only needs to process relatively similar data, thereby reducing the complexity of the model, making the model training faster and the prediction performance better. 3. Improving the generalization ability of the model and reducing the risk of overfitting: Each fault diagnosis model only needs to capture the vibration characteristics under a single working condition, rather than all vibration characteristics under all working conditions. Establishing a fault diagnosis model under specific working conditions helps reduce the model's sensitivity to noise and the risk of overfitting.

[0053] In the above, vibration detection and operating status detection are performed on the refrigeration compressor of the refrigeration equipment under complex working conditions to obtain vibration signals and working condition signals, and the working conditions are divided and fault diagnosis is performed based on the pre-processed vibration characteristic data and working condition data, thereby eliminating the interference between multiple vibration characteristics between different working conditions, reducing the misjudgment of normal vibration characteristics between different working conditions, and improving the accuracy of fault diagnosis; and, based on the historical healthy vibration data of each working condition, a fault diagnosis model is constructed and trained, and the vibration characteristic data under each working condition is diagnosed abnormally through the trained fault diagnosis model. Only positive sample training is used in model training, which can avoid the problem of low accuracy due to the lack of abnormal fault samples, and is more conducive to the inspection and maintenance of refrigeration equipment.

[0054] The following describes the various steps of the refrigeration equipment fault prediction and diagnosis method.

[0055] In step S1, the vibration signal is collected by a triaxial vibration sensor installed at a key position of a refrigeration device such as a refrigeration compressor, which can collect the vibration signal of each key component more accurately, so that the noise of the original vibration signal is small. Figure 3 As shown, a key position is a component or special position that is prone to abnormal vibration and easily causes refrigeration equipment failure. Optionally, a key position is one or more of the following: motor drive end A1, gear box input shaft A2, primary screw input end A3, primary screw output end A4, secondary screw input end A5 and secondary screw output end A6; the tested components may include motors, bearings, gears, transmission shafts, male and female screws and other components.

[0056] The working condition signal is collected by the working condition sensor installed at the second key position of the refrigeration equipment, and the second key position of the refrigeration equipment includes one or more of the following: pressure valve, exhaust port, air inlet, bearing and other positions. The working condition sensor can be a temperature sensor, flow sensor and other sensors that come with the refrigeration compressor. In this embodiment, the working condition signal is described by taking the pressure signal and temperature signal of the refrigeration compressor as an example.

[0057] Step S1 may include the following steps:

[0058] S11: Collect vibration signals and operating signals of key components of refrigeration equipment under various operating conditions, and use Z-score normalization method to normalize the characteristic data of vibration signals and the original data of operating signals, respectively, to obtain vibration characteristic time domain series and operating condition data time domain series.

[0059] S12: Filter out the vibration feature values ​​in the vibration feature time domain sequence whose absolute values ​​are greater than the vibration feature threshold, then determine the maximum vibration feature value and the minimum vibration feature value based on the current vibration feature time domain sequence, and determine the upper and lower limits of re-filtering, and filter out the vibration feature values ​​in the current vibration feature time domain sequence that are greater than the upper limit or less than the lower limit to obtain valid vibration feature data.

[0060] In this embodiment, the characteristic data of the vibration signal may include: vibration severity, kurtosis, high frequency root mean square, etc. The vibration characteristic threshold may correspond to the vibration severity threshold, the kurtosis threshold, and the high frequency root mean square threshold.

[0061] In order to reduce the interference of noise in vibration signals and operating condition signals on refrigeration equipment fault diagnosis, the interference noise is removed before data analysis and processing through the algorithm model. Z-score standardization can filter out each point that exceeds the threshold as invalid data to ensure the reliability and validity of the data.

[0062] Taking kurtosis as an example, the Z-score standardization method is used for noise reduction. The specific steps may include:

[0063] S121: Using formula (1) Calculate the Z-score kurtosis value.

[0064] In formula (1) is the mean value of the selected vibration signal, σ l is the standard deviation of the selected vibration signal, l m is the selected vibration characteristic time domain sequence, that is, X kurtosis , Y kurtosis , Z kurtosis Any signal in kurtosis , Y kurtosis , Z kurtosisThe vibration signal of the refrigeration equipment in the X, Y, and Z directions collected by the three-axis vibration sensor is calculated and obtained. zscore Represents X kurtosis , Y kurtosis , Z kurtosis The Z-score kurtosis value of any signal in after Z-score normalization.

[0065] The sampling frequency of the three-axis vibration sensor is 26000 Hz. The three-axis vibration sensor obtains the vibration signals of the X-axis, Y-axis and Z-axis of the refrigeration compressor over a period of time. The time domain sequences of the vibration signals of the X-axis, Y-axis and Z-axis are recorded as X, Y and Z, respectively. vibrate ={x1, x2, ..., x N},Y vibrate ={y1, y2, ..., y N}, Z vibrate ={z1, z2, ..., z N}.

[0066] X vibrate , Y vibrate and Z vibrate The system is divided into M time periods of equal length, and various vibration characteristic time domain sequences along the three axes are calculated respectively, such as vibration intensity, high-frequency root mean square, kurtosis, etc.

[0067] Optionally, the formula (2) Calculate the kurtosis of the time domain series X kurtosis , Y kurtosis and Z kurtosis .

[0068] Optionally, the formula (3) Calculate the RMS time-domain series X RMS , Y RMS and Z RMS .

[0069] In the above two formulas, M is the number of time periods, N is the number of sample points, and k is m For X kurtosis , Y kurtosis and Z kurtosis Any signal in X, a(m) is vibrate , Y vibrate and Z vibrate Any signal in a is the standard of a(m), is the average value of a(m); r m For X RMS , Y RMS and Z RMS Any signal in .

[0070] S122: Multiple Z-score kurtosis values ​​S zscore Composed of a new kurtosis time domain sequence K zscore (m).

[0071] S123: K zscore The vibration characteristic data whose absolute value is greater than the kurtosis threshold in (m) are filtered out to obtain k new (v).

[0072] S124: K new Substitute the maximum and minimum values ​​of (v) into formula (4):

[0073] k upper =k new_max (v)+(k new_max (v)-k new_min (v))*s to get the corresponding filtering upper limit k upper In formula (4), k new_max (v) is k new The maximum value of (v), k new_min (v) is k new (v) is the minimum value, s is the upper and lower limit expansion percentage, and the default value is 0.1.

[0074] S125: K new Substitute the maximum and minimum values ​​of (v) into formula (5):

[0075] k lower =k new_min (v)-(k new_max (v)-k new_min (v))*s to get the corresponding filtering lower limit k lower In formula (5), k new_min (v) is k new (v) is the minimum value, s is the upper and lower limit expansion percentage, and the default value is 0.1.

[0076] S126: The current kurtosis time domain sequence K zscore (m) is greater than the upper limit k upper Or less than the lower limit k lower The vibration characteristic values ​​are filtered out to obtain valid vibration characteristic data.

[0077] S13: Filter out the data whose absolute value in the time domain sequence of the operating condition data is greater than the operating condition data threshold, then determine the maximum operating condition data value and the minimum operating condition data value based on the current time domain sequence of the operating condition data, and determine the upper and lower limits of the re-filtering, and filter out the operating condition data greater than the upper limit or less than the lower limit in the current time domain sequence of the operating condition data to obtain valid operating condition data.

[0078] The specific steps of this embodiment can refer to the various steps of denoising using the Z-score normalization method taking kurtosis as an example in step S12, which will not be described in detail here.

[0079] In step S2, when aligning the vibration feature data and the working condition data, the data alignment can be performed based on the same timestamp. For example, if the vibration feature data is packaged every 15 minutes and the working condition data is packaged every 5 minutes, the vibration feature data can be used as the main data, and the working condition data closest to it can be found for data alignment, so that the two data are consistent in data dimension.

[0080] After the vibration characteristic data and the operating condition data are aligned, the operating conditions are divided according to the operating condition data, and then the vibration characteristic data under each operating condition are classified to obtain multiple groups of fault diagnosis data sets, thereby reducing the complexity of the vibration characteristic data and the difficulty of data processing, which helps to improve the fault diagnosis accuracy of refrigeration equipment.

[0081] The working condition division can be based on the change of working condition data. The working condition signal is the intake pressure signal, exhaust pressure signal, intake temperature signal, and any signal in the exhaust temperature signal that drives the refrigeration equipment to operate, and the working condition is divided respectively. Specifically, if the working condition signal is an intake pressure signal, the intake pressure signal can be divided into primary pressure, secondary pressure, and tertiary pressure according to the pressure intensity. The primary pressure is working condition one, the secondary pressure is working condition two, and the tertiary pressure can be working condition three. Then, based on the timestamp, the vibration characteristic data corresponding to working condition one, working condition two, and working condition three are found and classified. If the working condition signal is an intake temperature signal, the intake temperature signal can be divided into primary temperature, secondary temperature, and tertiary temperature according to the temperature. The primary temperature is working condition one, the secondary temperature is working condition two, and the tertiary temperature can be working condition three. Then, based on the timestamp, the vibration characteristic data corresponding to working condition one, working condition two, and working condition three are found and classified.

[0082] When the change of working condition data is small, it is difficult to divide the working conditions using the above method. At this time, the working condition division can be achieved through the isolation forest algorithm model.

[0083] In step S3, the fault diagnosis model is an improved OCSVM model. The OCSVM model is an unsupervised learning model that can be trained based on a single sample (positive sample), namely, historical healthy vibration data, thereby avoiding low diagnostic accuracy due to a lack of abnormal vibration data.

[0084] like Figure 4 As shown, step S3 may include four steps from step S31 to step S34.

[0085] S31: Based on each working condition, the historical healthy vibration data of the refrigeration equipment is obtained, the initial OCSVM model is trained based on the historical healthy vibration data, the optimal dictionary matrix and the sparse representation matrix are determined, the vibration feature data to be measured of the corresponding working condition is input into the initial OCSVM model, and the sparse representation matrix is ​​updated based on the vibration feature data to be measured. The multi-core hybrid kernel function of the initial OCSVM model is obtained by fusion of the Gaussian kernel function and the linear kernel function.

[0086] The historical healthy vibration data is used as the positive sample of vibration characteristics, which can be any group of data in a certain vibration characteristic data, for example, it can be X kurtosis , Y kurtosis and Z kurtosis Any set of data can be X in the high-frequency RMS RMS , Y RMS and Z RMS Any set of data.

[0087] Specifically, step S31 is implemented by the following steps:

[0088] S311: Based on each working condition, an initial OCSVM model is obtained, and the optimal dictionary matrix D and sparse representation matrix A are solved using the first optimization function. The first optimization function is:

[0089] Among them, D is the dictionary matrix, the size is d×k data matrix, where d is the data dimension, k is the number of samples, X is the input vector, that is, the training sample, that is, the size is d×n data matrix composed of historical healthy vibration data under various working conditions, d is the data dimension, n is the number of samples, A is the sparse representation matrix, which represents the sparse representation of each sample on the dictionary, the size is k×n, DA is the corresponding output vector, ||X-DA|| F represents the Frobenius norm, which is used to represent the reconstruction error between X and DA, and λ is the sparsity parameter, which is used to control the sparsity of the sparse representation.

[0090] By minimizing this optimization function, a set of coefficients A can be obtained so that the product of the input and the dictionary matrix D is as close to the output as possible, while limiting the size of the dictionary matrix D to avoid overfitting.

[0091] The principle of the first optimization function is to find the sparse representation matrix A by minimizing the reconstruction error between the input data (training sample X) and DA, where X≈D·A.

[0092] S312: Based on each working condition, the corresponding vibration characteristic data to be tested is input into the initial OCSVM model, and the corresponding vibration characteristic data to be tested is used as the sample x to be tested. i , use the second optimization function to solve the sparse representation matrix ai and update.

[0093] The second optimization function is Among them, x i is the input vector, i.e. the sample to be tested, Da i is the corresponding output vector; a i is a sparse representation matrix, representing the sample x to be tested i Sparse representation on the optimal dictionary matrix D, size k, ||x i -Da i ||2 is the L2 norm, which is used to represent the sample x to be tested. i and Da i The reconstruction error between .

[0094] By minimizing this optimization function, we can get a set of coefficients a i , so that the product of the input and the dictionary matrix D is as close to the output as possible, while limiting the size of the dictionary matrix D to avoid overfitting.

[0095] The principle of the second optimization function is: by minimizing the input data (test sample x i ) and Da i The reconstruction error between finds the sparse representation matrix a i , a i Replace A to implement the update of the sparse representation matrix.

[0096] In this embodiment, the sparse representation learning of the OCSVM model has the following advantages. First, the feature extraction and representation capabilities are powerful: refrigeration compressors usually have complex vibration characteristics and process signal characteristics. Sparse representation learning can extract the most representative features from the original data and represent the samples through sparse coefficients, thereby improving the expressiveness and robustness of the features. Second, it can adapt to different data distributions: The vibration data and process data of refrigeration compressor equipment may have different distribution characteristics. Sparse representation learning can dynamically learn dictionaries and sparse representations according to the distribution of the data, thereby better adapting to different data distributions.

[0097] S32: Based on the multi-core hybrid kernel function, the initial OCSVM model is reconstructed through the optimal dictionary matrix and the updated sparse representation matrix to obtain the fault diagnosis model.

[0098] Specifically, step S32 is implemented by the following steps:

[0099] S321: Determine the Gaussian kernel function as

[0100] Among them, (x i , x j ) is the data matrix composed of the optimal dictionary matrix and the updated sparse representation matrix, xi is the i-th data in the data matrix, x j is the jth data of the data matrix, ||x i -x j || is data x i and data x j The Euclidean distance between them, σ is the bandwidth parameter of the Gaussian kernel function, which is used to control the smoothness of the kernel function and can be determined by cross-validation. The Gaussian kernel function calculates the input data x i and x j The Gaussian distance between them captures the nonlinear characteristics of the data.

[0101] S322: Determine the linear kernel function as

[0102] Among them, (x i , x j ) is the data matrix composed of the optimal dictionary matrix and the updated sparse representation matrix, x i is the i-th data in the data matrix, x j is the jth data of the data matrix, T is the transpose operator, which is used to transpose the vector so that the inner product operation can be performed. i Transpose it to a row vector and multiply it with the column vector to get a scalar, which is the inner product of the two vectors. The linear kernel function calculates the input data x i and x j The inner product between them captures the linear relationship of the input data.

[0103] S323: Setting a weight parameter α of the Gaussian kernel function and the linear kernel function, α is greater than 0 and less than 1, and based on the weight parameter α, the Gaussian kernel function and the linear kernel function are merged to obtain a multi-core hybrid kernel function of the fault diagnosis model, the multi-core hybrid kernel function is:

[0104] K(x i ,x j )=αK Gaussian (x i ,x j )+(1-α)K Linear (x i ,x j );

[0105] S324: Determine the decision boundary, decision boundary parameter w and offset b of the fault diagnosis model based on the multi-core hybrid kernel function.

[0106] The decision boundary refers to an area containing normal data, usually represented as a hypersphere or a complex shape. The specific shape depends on the kernel function used. The decision boundary of this embodiment depends on the mixed kernel function after the fusion of the Gaussian kernel function and the linear kernel function, which is used to separate the normal vibration feature data from the abnormal vibration feature data.

[0107] In this embodiment, multi-core kernel function fusion has the following advantages. First, more comprehensive feature expression: the vibration data and process data of the refrigeration compressor may have diverse characteristics. Multi-core fusion can combine the advantages of different kernel functions to provide more comprehensive and diverse feature expression capabilities, thereby better capturing the complexity and diversity of the data. Second, enhance the robustness of the model: multi-core fusion can reduce the model's dependence on a single kernel function, making the model more robust and able to adapt to vibration and process data of different data types and distributions.

[0108] S33: Based on the decision boundary of the fault diagnosis model, the abnormality detection score of each vibration feature data is calculated, and the preset initial abnormality detection threshold is adjusted according to the abnormality detection score to obtain a dynamic threshold, wherein the abnormality detection score is the distance between the vibration feature data and the decision boundary of the fault diagnosis model.

[0109] Specifically, step S33 is implemented by the following steps:

[0110] S331: Input the vibration characteristic data to be measured corresponding to the working condition into the OCSVM model, and calculate the abnormal detection score of each vibration characteristic data based on the decision boundary, decision boundary parameter w and offset b of the OCSVM model. The calculation formula of the abnormal detection score is: Among them, D istance (x i ) is the anomaly detection score, φ(x i ) is a multi-core hybrid kernel function, which is used to map the input data from the original space to the high-dimensional feature space. T is the transpose operator, which is used to transpose the vector. w is the decision boundary parameter, b is the offset, and ||w|| is the modulus of the decision boundary parameter w.

[0111] S332: Calculate the threshold adjustment factor according to the anomaly detection score and the preset initial anomaly detection threshold. The calculation formula of the threshold adjustment factor is: Among them, β i is the threshold adjustment factor, ρ0 is the preset initial anomaly detection threshold;

[0112] S333: Determine a dynamic threshold according to a preset initial abnormality detection threshold, a threshold adjustment factor, and a preset threshold adjustment speed. The calculation formula of the dynamic threshold is:

[0113] ρ i=ρ0+y·(β i -1)·ρ0, where ρ i is the dynamic threshold, and γ is the learning rate parameter, which is used to control the speed of threshold adjustment.

[0114] In this embodiment, the use of dynamic thresholds for abnormal point determination has the following advantages. First, it can adapt to data changes in real time: the vibration data and process data of the refrigeration compressor may change with time and operating conditions. The dynamic threshold adjustment can dynamically adjust the threshold of abnormal detection according to the changes in real-time data, so that the abnormal detection algorithm can adapt to the changes in data in real time, improving the real-time and accuracy of the algorithm. Second, it can reduce the false alarm rate: by dynamically detecting and adjusting the threshold, the abnormal detection threshold can be flexibly fine-tuned according to the distribution of the current data, thereby reducing the false alarm rate and improving the accuracy and reliability of abnormal detection.

[0115] S34: The vibration feature data with an abnormal detection score greater than the dynamic threshold is regarded as an abnormal fault point.

[0116] In this embodiment, the dynamic threshold is ρ i , sample x i The anomaly detection score is When the anomaly detection score meets the condition Distance(x i )>ρ i , then the sample x i are considered as outliers.

[0117] like Figure 3 and 5 As shown, Figure 5 The left side in the middle is the vibration sensor characteristic value selection column, which specifically includes the vibration characteristic data and operating condition data input when building the abnormal diagnosis model. Figure 3 In the figure, the vibration data measurement points (measurement point locations) include the motor drive end A1, the gearbox input shaft A2, the primary screw input end A3, the primary screw output end A4, the secondary screw input end A5, and the secondary screw output end A6. The corresponding characteristic values ​​can be selected as vibration characteristic data and input into the abnormal diagnosis model. The working condition data measurement points include the secondary intake pressure, the secondary exhaust pressure, the secondary intake temperature, and the secondary exhaust temperature. The right side of the figure is a schematic diagram of abnormal diagnosis through the abnormal diagnosis model, where the blue dots are normal points and the red dots are abnormal points.

[0118] In step S4, for the abnormal fault point under each working condition, a spectrum analysis method is used to perform fault location analysis, to obtain fault location information corresponding to each working condition, and to generate and display a maintenance work order for the refrigeration equipment.

[0119] Specifically, step S4 may include four steps from step S41 to step S44.

[0120] S41: determining the time domain waveform of the corresponding vibration signal for each abnormal fault point under each working condition, and converting the time domain waveform of the vibration signal into a frequency domain waveform of the vibration signal;

[0121] S42: determining a specific frequency of the frequency domain waveform of the vibration signal, and then determining an amplitude of the specific frequency, wherein the amplitude of the specific frequency corresponds to a refrigeration equipment fault;

[0122] S43: determining fault location information according to the amplitude change of a specific frequency in the frequency domain waveform;

[0123] S44: predicting the cause of the fault according to the fault location information and generating maintenance suggestions, and generating a maintenance work order display for the refrigeration equipment based on the fault location information, the predicted cause of the fault and the maintenance suggestions.

[0124] Based on experience, after obtaining the fault location information, the cause of the fault can be predicted, and the most likely cause of the fault can be displayed on the maintenance work order, so that maintenance personnel can check and repair the equipment in a targeted manner. The maintenance work order is sent from the server to the client terminal for display, such as on the APP sent to the user's mobile phone, so that the user can check the fault information of the refrigeration equipment in time and take timely measures to repair the refrigeration equipment.

[0125] In an optional embodiment, in order to make the fault prediction diagnosis of refrigeration equipment more accurate, the fault diagnosis model and the anomaly detection threshold are continuously corrected in the form of maintenance work order feedback, and the fault diagnosis model is iteratively updated, so that the algorithm model has stronger generalization ability, and the fault prediction accuracy continues to improve with the passage of time, thereby further improving the accuracy of fault diagnosis and early warning. Specifically, the dictionary matrix, decision boundary, and anomaly detection threshold of the fault diagnosis model can be updated.

[0126] Different work order information will be generated according to different refrigeration equipment failures. The feedback information on the maintenance work order may include a variety of feedback information, such as bearing lubrication information, confirmation information on whether the bearing is faulty during inspection, the last maintenance time, and the last maintenance item information.

[0127] Optionally, in order to update and iterate the abnormal diagnosis model, after step S4, step S5 is also included to receive the user's feedback information on the maintenance work order to update the healthy vibration data of the refrigeration equipment, iteratively train the fault diagnosis model through the updated healthy vibration data, and continuously correct the abnormal detection threshold of the fault diagnosis model in multiple maintenance work order feedbacks, so as to make the abnormal diagnosis model more accurate. The abnormal detection threshold is used to determine whether the currently input vibration characteristic data is an abnormal fault point.

[0128] During the model iteration process, the healthy vibration data is continuously updated, and the dictionary matrix and sparse representation matrix of the OCSVM model are also continuously updated. The fused multi-core hybrid kernel function is updated synchronously, so that the decision boundary of the OCSVM model can be adjusted. The decision boundary is closer to the boundary between normal vibration data and abnormal vibration data, so that the fault diagnosis model can detect abnormal fault points more accurately and reduce the false alarm rate.

[0129] In the above technical scheme of the present embodiment, vibration detection and operating status detection are performed on the refrigeration compressor of the refrigeration equipment under complex working conditions to obtain vibration signals and working condition signals, and the working conditions are divided and fault diagnosis is performed based on the pre-processed vibration characteristic data and working condition data, thereby eliminating the interference between multiple vibration characteristics between different working conditions, reducing the misjudgment of normal vibration characteristics between different working conditions, and improving the accuracy of fault diagnosis; and, based on the historical healthy vibration data of each working condition, a fault diagnosis model is constructed and trained, and the vibration characteristic data under each working condition is diagnosed abnormally through the trained fault diagnosis model. Only positive sample training is used in model training, which can avoid the problem of low accuracy due to the lack of abnormal fault samples, and is more conducive to the inspection and maintenance of refrigeration equipment.

[0130] Furthermore, the fault diagnosis model is obtained by improving the OCSVM model, which can continue the advantages of the OCSVM model to improve the feature expression ability, enhance the model robustness, and better adapt to the data distribution of different working conditions; at the same time, the fault diagnosis model based on multi-core function fusion can express features more comprehensively, provide more comprehensive and diversified feature expression capabilities, and can adapt to vibration and working condition data distributed in different data types, reduce the fault diagnosis model's dependence on a single function, and make the model more robust.

[0131] Furthermore, when diagnosing abnormal fault points, fixed threshold comparison is abandoned and dynamic threshold comparison is adopted, which can enable the model to better adapt to the real-time changes of the vibration data and operating data of the refrigeration compressor, thereby improving the real-time performance and accuracy of the algorithm; the dynamic threshold can be flexibly adjusted to reduce the false alarm rate.

[0132] Furthermore, the healthy vibration data of the refrigeration equipment can be updated through the feedback information of the maintenance work order, and the fault diagnosis model can be iteratively trained through the updated healthy vibration data, and the abnormal detection threshold can be continuously corrected, so that the fault diagnosis model can predict the fault point more accurately as the use time goes by.

[0133] like Figure 6 As shown, Figure 6 A schematic diagram of the structure of a refrigeration equipment fault prediction and diagnosis device provided in an embodiment of the present application.

[0134] The refrigeration equipment fault prediction and diagnosis device includes:

[0135] The signal acquisition and processing module 201 is used to acquire the vibration signals and working condition signals of each key component of the refrigeration equipment under various working conditions, and pre-process the original vibration characteristic data in the vibration signal and the original working condition data in the working condition signal to obtain the vibration characteristic data and working condition data;

[0136] The working condition classification module 202 is used to align the vibration characteristic data with the working condition data, classify the working conditions according to the working condition data, and classify the vibration characteristic data under each working condition to obtain multiple groups of fault diagnosis data sets;

[0137] The abnormality detection module 203 is used to obtain the historical healthy vibration data of the refrigeration equipment based on each working condition, use the historical healthy vibration data and the vibration characteristic data to be tested to build and train a fault diagnosis model, perform abnormal diagnosis on the vibration characteristic data of the corresponding working condition through the trained fault diagnosis model, and obtain abnormal fault points, wherein the fault diagnosis model is obtained by improving the OCSVM model, and the kernel function of the fault diagnosis model is a multi-core hybrid kernel function obtained by fusing a Gaussian kernel function and a linear kernel function;

[0138] The fault diagnosis module 204 is used to perform fault location analysis for the abnormal fault point under each working condition by using a spectrum analysis method, obtain fault location information corresponding to each working condition, generate a maintenance work order for the refrigeration equipment, and display it.

[0139] In an optional embodiment, a model correction module is also included, which is used to receive user feedback information on the maintenance work order to update the healthy vibration data of the refrigeration equipment, iteratively train the fault diagnosis model through the updated healthy vibration data, and continuously correct the abnormal detection threshold of the fault diagnosis model through multiple maintenance work order feedbacks. The abnormal detection threshold is used to determine whether the currently input vibration characteristic data is an abnormal fault point.

[0140] In an optional embodiment, the anomaly detection module 203 includes:

[0141] A data acquisition unit is used to acquire historical healthy vibration data of the refrigeration equipment based on each working condition, use the historical healthy vibration data to train an initial OCSVM model, determine an optimal dictionary matrix and a sparse representation matrix, input the vibration characteristic data to be measured corresponding to the working condition into the initial OCSVM model, and update the sparse representation matrix based on the vibration characteristic data to be measured, wherein the multi-core hybrid kernel function of the initial OCSVM model is obtained by fusion of a Gaussian kernel function and a linear kernel function;

[0142] A model reconstruction unit, used to reconstruct the initial OCSVM model through the optimal dictionary matrix and the updated sparse representation matrix based on a multi-core hybrid kernel function to obtain a fault diagnosis model;

[0143] A threshold adjustment unit, configured to calculate an abnormality detection score for each of the vibration characteristic data based on a decision boundary of the fault diagnosis model, and to adjust an initial abnormality detection threshold according to the abnormality detection score to obtain a dynamic threshold;

[0144] The fault point detection unit is used to take the vibration characteristic data with the abnormal detection score greater than the dynamic threshold as an abnormal fault point.

[0145] The refrigeration equipment fault prediction and diagnosis device of this embodiment is a device produced by the same inventive concept as the refrigeration equipment fault prediction and diagnosis method. The contents not recorded in this embodiment can refer to the contents recorded in the above-mentioned refrigeration equipment fault prediction and diagnosis method, and solve the same technical problems and obtain the same technical effects.

[0146] The present application also provides a refrigeration equipment fault prediction and diagnosis system. Figure 1 As shown, the refrigeration equipment fault prediction and diagnosis system 100 includes: a vibration sensor 101, which is used to collect vibration signals of the refrigeration equipment under various working conditions; an operating condition sensor 102, which is used to collect operating condition signals of the refrigeration equipment under various working conditions; a server 103, which is used to perform predictive fault diagnosis on the refrigeration equipment based on the vibration signal and the operating condition signal, using the refrigeration equipment fault prediction and diagnosis method described in the above embodiment.

[0147] In an embodiment of the present application, vibration detection and operating status detection are performed on the refrigeration compressor of the refrigeration equipment under complex working conditions to obtain vibration signals and working condition signals, and the working conditions are divided based on the preprocessed vibration characteristic data and working condition data, and fault diagnosis is performed jointly, thereby eliminating interference between multiple vibration characteristics between different working conditions, reducing misjudgment of normal vibration characteristics between different working conditions, and improving the accuracy of fault diagnosis; and, based on the historical healthy vibration data of each working condition, a fault diagnosis model is constructed and trained, and the vibration characteristic data under each working condition is diagnosed abnormally through the trained fault diagnosis model. Only positive sample training is used in model training, which can avoid the problem of low accuracy due to the lack of abnormal fault samples, and is more conducive to the inspection and maintenance of refrigeration equipment.

[0148] Note that the above are only preferred embodiments of the present application and the technical principles used. Those skilled in the art will understand that the present application is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for predicting and diagnosing refrigeration equipment faults, characterized in that: The steps include: Collect vibration signals and working condition signals of key components of refrigeration equipment under various working conditions, pre-process original vibration characteristic data in vibration signals and original working condition data in working condition signals, and obtain vibration characteristic data and working condition data; Aligning the vibration characteristic data with the operating condition data, dividing the operating condition according to the operating condition data, and classifying the vibration characteristic data based on each operating condition to obtain multiple groups of fault diagnosis data sets; Based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, and a fault diagnosis model is constructed and trained using the historical healthy vibration data and the vibration characteristic data to be tested, and abnormal diagnosis is performed on the vibration characteristic data of the corresponding working condition through the trained fault diagnosis model to obtain abnormal fault points; For the abnormal fault point under each working condition, a spectrum analysis method is used to perform fault location analysis, to obtain fault location information corresponding to each working condition, and to generate and display a maintenance work order for the refrigeration equipment; Wherein, based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, the historical healthy vibration data and the vibration characteristic data to be tested are used to build and train a fault diagnosis model, and the vibration characteristic data of the corresponding working condition is abnormally diagnosed by the trained fault diagnosis model to obtain the abnormal fault point, including: Based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, the historical healthy vibration data is used to train an initial OCSVM model, an optimal dictionary matrix and a sparse representation matrix are determined, vibration feature data to be measured corresponding to the working condition is input into the initial OCSVM model, and the sparse representation matrix is ​​updated based on the vibration feature data to be measured, wherein the multi-core hybrid kernel function of the initial OCSVM model is obtained by fusion of a Gaussian kernel function and a linear kernel function; Based on a multi-core hybrid kernel function, reconstructing an initial OCSVM model through the optimal dictionary matrix and the updated sparse representation matrix to obtain a fault diagnosis model; Based on the decision boundary of the fault diagnosis model, calculating the abnormality detection score of each vibration characteristic data, and adjusting the initial abnormality detection threshold according to the abnormality detection score to obtain a dynamic threshold; The vibration characteristic data whose abnormal detection score is greater than the dynamic threshold is taken as an abnormal fault point.

2. The refrigeration equipment fault prediction and diagnosis method according to claim 1, characterized in that: Based on each working condition, historical healthy vibration data of the refrigeration equipment is obtained, the historical healthy vibration data is used to train the initial OCSVM model, an optimal dictionary matrix and a sparse representation matrix are determined, vibration feature data to be measured corresponding to the working condition is input into the initial OCSVM model, and the sparse representation matrix is ​​updated based on the vibration feature data to be measured, including: Based on each working condition, an initial OCSVM model is obtained, and the optimal dictionary matrix D and sparse representation matrix A are solved using the first optimization function. The first optimization function is: Among them, D is the dictionary matrix, the data matrix of size d×k, where d is the data dimension and k is the number of samples, X is the training sample, that is, the data matrix of size d×n composed of the historical healthy vibration data under each working condition, d is the data dimension, n is the number of samples, A is the sparse representation matrix, which represents the sparse representation of each sample on the dictionary, with a size of k×n, ||X-DA|| F represents the Frobenius norm, which is used to represent the reconstruction error between X and DA, and λ is the sparsity parameter, which is used to control the sparsity of the sparse representation; Based on each working condition, the corresponding vibration characteristic data to be tested is input into the initial OCSVM model, and the corresponding vibration characteristic data to be tested is used as the sample x to be tested. i , use the second optimization function to solve the sparse representation matrix a i And update, the second optimization function is Among them, a i is a sparse representation matrix, representing the sample x to be tested i Sparse representation on the optimal dictionary matrix D, size k, ||x i -Da i ||2 represents the L2 norm, which is used to represent the sample x to be tested i and Da i The reconstruction error between .

3. The refrigeration equipment fault prediction and diagnosis method according to claim 2, characterized in that: The method of reconstructing the initial OCSVM model based on the multi-core hybrid kernel function through the optimal dictionary matrix and the updated sparse representation matrix to obtain a fault diagnosis model includes: Determine the Gaussian kernel function as Among them, (x i , x j ) is the data matrix composed of the optimal dictionary matrix and the updated sparse representation matrix, x i is the i-th data in the data matrix, x j is the jth data of the data matrix, σ is the bandwidth parameter of the Gaussian kernel function; Determine the linear kernel function as Among them, (x i , x j ) is the data matrix composed of the optimal dictionary matrix and the updated sparse representation matrix, x i is the i-th data in the data matrix, x j is the jth data of the data matrix, T is the transposition operator, which is used to transpose the vector; The weight parameter α of the Gaussian kernel function and the linear kernel function is set, α is greater than 0 and less than 1, and the Gaussian kernel function and the linear kernel function are fused based on the weight parameter α to obtain the multi-core hybrid kernel function of the fault diagnosis model. The multi-core hybrid kernel function is: K(x i ,x j )=αK Gaussian (x i ,x j )+(1-α)K Linear (x i ,x j ); A decision boundary of a fault diagnosis model is determined based on the multi-core hybrid kernel function.

4. The refrigeration equipment fault prediction and diagnosis method according to claim 3, characterized in that: The decision boundary based on the fault diagnosis model, calculating the abnormality detection score of each vibration feature data, and adjusting the initial abnormality detection threshold according to the abnormality detection score to obtain a dynamic threshold, comprises: Based on the decision boundary of the fault diagnosis model, the abnormality detection score of each vibration feature data is calculated. The calculation formula of the abnormality detection score is: Among them, Distance(x i ) is the anomaly detection score, φ(x i ) is a multi-core hybrid kernel function, T is a transpose operator used to transpose a vector, w is a decision boundary parameter, b is an offset, and ||w|| is the modulus of the decision boundary parameter w; The threshold adjustment factor is calculated according to the anomaly detection score and the preset initial anomaly detection threshold. The calculation formula of the threshold adjustment factor is: Among them, β i is the threshold adjustment factor, ρ0 is the preset initial anomaly detection threshold; The dynamic threshold is determined according to the preset initial abnormality detection threshold, the threshold adjustment factor, and the preset threshold adjustment speed. The calculation formula of the dynamic threshold is ρ i =ρ0+γ·(β i -1)·ρ0, where ρ i is the dynamic threshold, and γ is the learning rate parameter, which is used to control the speed of threshold adjustment.

5. The refrigeration equipment fault prediction and diagnosis method according to claim 1, characterized in that: The method of using a spectrum analysis method to perform fault location analysis on the abnormal fault point under each working condition, obtains fault location information corresponding to each working condition, generates and displays a maintenance work order for the refrigeration equipment, including: Determine the time domain waveform of the corresponding vibration signal for each abnormal fault point under each working condition, and convert the time domain waveform of the vibration signal into a frequency domain waveform of the vibration signal; Determine a specific frequency of the frequency domain waveform of the vibration signal, and then determine the amplitude of the specific frequency, wherein the amplitude of the specific frequency corresponds to a refrigeration equipment fault; Determining fault location information based on the amplitude change of a specific frequency in the frequency domain waveform; The cause of the fault is predicted according to the fault location information and a maintenance suggestion is generated, and a maintenance work order display of the refrigeration equipment is generated based on the fault location information, the predicted cause of the fault and the maintenance suggestion.

6. The refrigeration equipment fault prediction and diagnosis method according to claim 1, characterized in that: Also includes the steps: Receive user feedback information on the maintenance work order to update healthy vibration data of the refrigeration equipment, iteratively train the fault diagnosis model through the updated healthy vibration data, and continuously correct the abnormal detection threshold of the fault diagnosis model in multiple maintenance work order feedbacks. The abnormal detection threshold is used to determine whether the currently input vibration characteristic data is an abnormal fault point.

7. The refrigeration equipment fault prediction and diagnosis method according to claim 1, characterized in that: The vibration signal is acquired by a vibration sensor installed on the refrigeration compressor; the operating condition signal is acquired by an operating condition sensor installed on the refrigeration compressor.

8. A refrigeration equipment fault prediction and diagnosis device, characterized in that: include: The signal acquisition and processing module is used to collect the vibration signals and working condition signals of each key component of the refrigeration equipment under various working conditions, and pre-process the original vibration characteristic data in the vibration signal and the original working condition data in the working condition signal to obtain the vibration characteristic data and working condition data; A working condition classification module, used for aligning the vibration characteristic data with the working condition data, classifying the working conditions according to the working condition data, and classifying the vibration characteristic data under each working condition to obtain multiple groups of fault diagnosis data sets; An abnormality detection module is used to obtain historical healthy vibration data of the refrigeration equipment based on each working condition, use the historical healthy vibration data and the vibration characteristic data to be tested to build and train a fault diagnosis model, and perform abnormal diagnosis on the vibration characteristic data of the corresponding working condition through the trained fault diagnosis model to obtain abnormal fault points; A fault diagnosis module is used to perform fault location analysis for the abnormal fault point under each working condition by using a spectrum analysis method, obtain fault location information corresponding to each working condition, generate a maintenance work order for the refrigeration equipment, and display it; Wherein, the anomaly detection module includes: A data acquisition unit is used to acquire historical healthy vibration data of the refrigeration equipment based on each working condition, use the historical healthy vibration data to train an initial OCSVM model, determine an optimal dictionary matrix and a sparse representation matrix, input the vibration characteristic data to be measured corresponding to the working condition into the initial OCSVM model, and update the sparse representation matrix based on the vibration characteristic data to be measured, wherein the multi-core hybrid kernel function of the initial OCSVM model is obtained by fusion of a Gaussian kernel function and a linear kernel function; A model reconstruction unit, used to reconstruct the initial OCSVM model through the optimal dictionary matrix and the updated sparse representation matrix based on a multi-core hybrid kernel function to obtain a fault diagnosis model; A threshold adjustment unit, configured to calculate an abnormality detection score for each of the vibration characteristic data based on a decision boundary of the fault diagnosis model, and to adjust an initial abnormality detection threshold according to the abnormality detection score to obtain a dynamic threshold; A fault point detection unit is used to take the vibration characteristic data whose abnormal detection score is greater than the dynamic threshold as an abnormal fault point; 9. Refrigeration equipment fault prediction and diagnosis system, characterized in that: include: Vibration sensor, used to collect vibration signals of refrigeration equipment under various working conditions; Working condition sensor, used to collect working condition signals of refrigeration equipment under various working conditions; The server is used to perform predictive fault diagnosis on the refrigeration equipment based on the vibration signal and the operating condition signal by using the refrigeration equipment fault prediction and diagnosis method according to any one of claims 1 to 8.

Images