A method for hydraulic piston pump fault diagnosis

Abstract

The application discloses a kind of methods for hydraulic plunger pump fault diagnosis, comprising the following steps: S1, plunger pump is tested by the standard test procedure set, then the key signal of outlet, oil inlet, oil return, shell surface, driving shaft is collected by data acquisition system;S2, extract the frequency domain feature and time domain feature in key signal;S3, correlation analysis is carried out to different key signals and related components, and correlation feature is extracted;S4, frequency domain feature, time domain feature and correlation feature are respectively scored by confidence degree using two ways of rule fault tree diagnosis model based on expert knowledge and fault classification model fusing expert knowledge and multi-source signal feature;S5, the confidence degree scoring of two different models is reordered, and the result with highest confidence degree is finally output as the final result output.The application has the characteristics that can effectively improve fault diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of hydraulic piston pump fault diagnosis technology, and in particular to a method for diagnosing hydraulic piston pump faults. Background Technology

[0002] Hydraulic pumps, especially hydraulic piston pumps, are widely used in many fields. Although hydraulic piston pumps offer excellent performance, their complex structure leads to high manufacturing and maintenance costs. Therefore, during production trials and other processes, it is necessary to diagnose and provide early warnings for hydraulic piston pump malfunctions, which can reduce costs and improve reliability.

[0003] Existing fault diagnosis technologies and processes are generally divided into three stages: First, key signals such as pressure, temperature, and speed of the pump are collected in real time by data acquisition equipment. Then, the data processing module performs data preprocessing such as sampling slice feature extraction on the real-time collected signal data. Finally, the extracted signal data features are classified into faults by a neural network model.

[0004] Current research on hydraulic pump fault diagnosis often focuses on analyzing typical faults of a particular pump type and then using a single sensor signal for diagnosis. Most studies employ a purely data-driven approach, failing to incorporate prior expert information, and the reliance on a single algorithm model can lead to low diagnostic accuracy. Furthermore, current fault diagnosis technologies mostly target specific high-frequency defects in their design and diagnosis, lacking comprehensive experimental procedures and effective signal feature extraction methods, which also negatively impacts diagnostic accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide a method for diagnosing faults in hydraulic piston pumps. This invention effectively improves the accuracy of fault diagnosis.

[0006] The technical solution of the present invention: a method for fault diagnosis of hydraulic piston pumps, comprising the following steps:

[0007] S1. The plunger pump is tested according to the established standard test procedure, and then the key signals of the oil outlet, oil inlet, oil return, housing surface, and drive shaft are collected through the data acquisition system.

[0008] S2. Extract frequency domain and time domain features from key signals;

[0009] S3. Perform correlation analysis on different key signals and related components, and extract correlation features;

[0010] S4. Confidence scores are assigned to frequency domain features, time domain features, and correlation features using two methods: a rule-based fault tree diagnostic model based on expert knowledge and a fault classification model that integrates expert knowledge and multi-source signal features, respectively.

[0011] S5. Reorder the confidence scores of the two different models and output the result with the highest confidence score as the final result.

[0012] In the aforementioned method for diagnosing faults in a hydraulic piston pump, the key signals include the pressure, flow rate, and temperature of the oil outlet, inlet, and return ports; the temperature of the housing surface; the vibration of the housing surface in the X, Y, and Z directions; and the torque and speed of the drive shaft.

[0013] In the aforementioned method for diagnosing faults in a hydraulic piston pump, the time-domain features include dimensional time-domain features and dimensionless time-domain features.

[0014] Dimensional time-domain characteristics include maximum value, maximum absolute value, minimum value, mean, peak-to-peak value, absolute mean, root mean square value, root mean square amplitude, and standard deviation.

[0015] Dimensionless time-domain features include kurtosis, skewness, peak value, impulse, margin, and waveform.

[0016] In the aforementioned method for diagnosing faults in a hydraulic piston pump, the frequency domain features include centroid frequency, average frequency domain amplitude, root mean square frequency, frequency variance, and maximum frequency component.

[0017] In the aforementioned method for fault diagnosis of hydraulic piston pumps, the rule-based fault tree diagnosis model based on expert knowledge constructs a fault label system based on expert knowledge and experience, and sets a fault tree process for each label; after extracting the time domain features, frequency domain features and correlation features of all signals, the fault tree process is used to analyze and diagnose each feature, and to make a fault type judgment and confidence score.

[0018] In the aforementioned method for fault diagnosis of hydraulic piston pumps, the fault classification model that integrates expert knowledge and multi-source signal features uses a CNN convolution model to perform multiple convolution and pooling operations on the acquired time-domain features, frequency-domain features, and correlation features to obtain feature vectors. Finally, the feature vectors are classified by fault category label using softmax, and the final confidence score is output.

[0019] Compared with existing technologies, this invention designs a standard experimental procedure and signal acquisition method based on industry expert knowledge. It performs feature extraction on the signal acquisition method and incorporates expert knowledge for correlation analysis during the feature extraction process. Simultaneously, this invention employs two different algorithm models for fault diagnosis (a rule-based fault tree diagnosis model based on expert knowledge and a fault classification model that integrates expert knowledge and multi-source signal features for confidence scoring), and uses a confidence re-ranking method across multiple algorithm models to calculate and output the final diagnostic results. This results in a fault diagnosis with high accuracy. In summary, this invention effectively improves the accuracy of fault diagnosis. Attached Figure Description

[0020] Appendix Figure 1 This is a flowchart illustrating the overall fault diagnosis process for the hydraulic plunger pump of this invention.

[0021] Appendix Figure 2 This invention relates to a rule-based fault diagnosis system based on expert knowledge.

[0022] Appendix Figure 3 This invention provides a fault classification model that integrates expert knowledge and multi-source signal characteristics.

[0023] Appendix Figure 4 The confidence levels of the diagnostic model of this invention are reordered. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0025] Example. A method for fault diagnosis of a hydraulic piston pump, comprising the following steps:

[0026] S1. The plunger pump is tested according to the established standard test procedure, and then the key signals of the oil outlet, oil inlet, oil return, housing surface, and drive shaft are collected through the data acquisition system.

[0027] S2. Extract frequency domain and time domain features from key signals;

[0028] S3. Perform correlation analysis on different key signals and related components, and extract correlation features;

[0029] S4. Confidence scores are assigned to frequency domain features, time domain features, and correlation features using two methods: a rule-based fault tree diagnostic model based on expert knowledge and a fault classification model that integrates expert knowledge and multi-source signal features, respectively.

[0030] S5. Reorder the confidence scores of the two different models and output the result with the highest confidence score as the final result.

[0031] Regarding the selection of a signal acquisition scheme for hydraulic pumps, hydraulic piston pumps exhibit a wide variety of failures, involving various components. When a hydraulic piston pump malfunctions, the pump's input, output, and related parameter information will change. Based on this, we first developed a standard signal acquisition scheme to collect pressure, flow, and temperature signals from the outlet, inlet, and return ports, as shown in Table 1.

[0032] Table 1. Hydraulic Pump Signal Acquisition Types

[0033]

[0034]

[0035] Based on the signal acquisition scheme and the operating characteristics of the hydraulic pump, this invention establishes a standardized signal acquisition process for hydraulic piston pumps. The aim is to standardize the sampling of data collected by sensors, thereby enabling standardized signal data feature extraction and fault diagnosis. The specific standard test procedure is shown in Table 2.

[0036] Table 2 Experimental Data Collection Flowchart

[0037]

[0038] Based on the standard test procedure and standard signal acquisition scheme, the time-domain characteristics, frequency-domain characteristics, and pump fault-related characteristics of different signals can effectively accommodate information about the pump's operating status. Furthermore, by analyzing the signal characteristics and the pump's sensitive parameters, this prior expert knowledge can be directly integrated into the fault diagnosis model. Taking vibration signals as an example, vibration signals have a high signal-to-noise ratio and a large amount of information, reflecting abnormal friction of internal moving parts, abnormal impact between components, and even abnormal conditions such as component cracking. Vibration signals contain a lot of information in both the time and frequency domains. Therefore, in this invention, feature extraction is performed on the 14 acquired signals, mainly including time-domain and frequency-domain features. Furthermore, expert knowledge is used to summarize which pump sensitive parameters these 14 signal features are related to, thereby formulating diagnostic rules and integrating them into the neural network diagnostic model. Specifically, the main features of the 14 signals are shown in Tables 3 to 5:

[0039] Table 3. Dimensional Time-Domain Characteristics

[0040]

[0041]

[0042] Table 4 Dimensionless Time Domain Characteristics

[0043]

[0044] Table 5 Frequency Domain Characteristics

[0045]

[0046] Based on 14 signal characteristics, this invention analyzes the structure, working mechanism, and failure modes of the entire pump, extracting features strongly correlated with pump failures. For example, the outlet / inlet pressure difference reflects the pressurization of the working medium by the hydraulic piston pump, indicating the overall efficiency of the hydraulic pump. The outlet / return port pressure difference can reflect the pump's efficiency while also assisting in monitoring leakage. Based on expert knowledge, this invention summarizes the correlation between various features and pump parameters, as follows:

[0047] Table 6. Relevant Characteristics of Hydraulic Piston Pumps

[0048]

[0049]

[0050] Based on the extracted signal features, this invention employs two main diagnostic methods: a rule-based fault tree diagnostic model based on expert knowledge, and a fault classification model that integrates expert knowledge with multi-source signal features. The final output is determined by ranking the confidence scores from both methods. The diagnostic model proposed in this application, which integrates multiple expert experiences and knowledge, can improve diagnostic accuracy.

[0051] The technical solution of the present invention will now be described in detail and completely with reference to the accompanying drawings.

[0052] refer to Figure 1 This invention provides a flowchart for the overall fault diagnosis of a hydraulic piston pump. The signal acquisition in this flowchart follows the standard test procedure given in Table 2 of this invention. Then, according to the signal acquisition scheme given in Table 1, 14 key signal indicators, such as pressure, temperature, and speed, are acquired from key components including the oil outlet, oil inlet, and oil return port. After the signals are acquired, frequency domain and time domain features of the 14 key signals are extracted according to the signal feature extraction schemes in Tables 3 to 5 of this invention. Furthermore, correlation analysis is performed on different signal features and related components based on expert knowledge to extract correlation features. Two methods are used for fault diagnosis based on the data features extracted from the signals. One method is to refer to… Figure 2 Rule-based fault diagnosis systems based on expert knowledge include, for example,... Figure 3 The fault classification model shown integrates expert knowledge and multi-source signal characteristics.

[0053] A rule-based fault diagnosis system based on expert knowledge constructs a fault labeling system based on experience, with each label having a fault tree process, referencing... Figure 2Taking valve spring failure as an example, a rule-based fault diagnosis process is presented. After extracting the different characteristics of 14 signals, the outlet pressure signal is first analyzed. Here, according to the standard test procedure, the outlet pressure is divided into two abnormal situations: all operating conditions and specific operating conditions (one or two labeled operating conditions). When the hydraulic pump outlet pressure is abnormal under all operating conditions, the average outlet flow rate is retrieved. If the average flow rate matches the set flow rate value of the labeled operating condition, it is considered to be a design problem of the hydraulic pump components, and parameter adjustment is required. If the average value does not match the set value, it is determined to be a failure of the hydraulic pump valve spring. If the spring is too stiff or too soft, the hydraulic pump flow-pressure characteristic curve needs to be plotted. The slope of the part of the curve that exceeds the set value is determined by calculating the slope. When k>k 原 When (where K is the stiffness coefficient of the fault spring, k) 原 (where k is the initial stiffness coefficient of the spring), indicating that the adjusting spring is too soft. <k 原 If the pressure is too stiff, the problem is identified as an excessively stiff pressure regulating spring. When the hydraulic pump's outlet pressure pulsates abnormally under specific operating conditions, the flow switching in the test procedure is reviewed. This involves examining the pressure signal characteristics during the transition between two different flow rate labels. If the amplitude is too high, the pulsation fundamental frequency is higher than normal, or high-order harmonics are present, the problem is considered to be excessive damping in the hydraulic pump's servo mechanism. If all these characteristics are normal, the problem is considered to be a fluid-structure interaction issue within the pump itself. The process of judging through flow adjustment described above is then repeated to determine the final fault type.

[0054] A fault classification model that integrates expert knowledge and multi-source signal characteristics, referencing Figure 3 The time-domain information, frequency-domain information, and correlation features of the 14 signals obtained in this invention are processed using a CNN convolutional model through multiple convolutional pooling operations to obtain feature vectors. Finally, softmax is used to classify the feature vectors into fault category labels. The entire network model is trained using labeled data. Label data is obtained from two sources: one is manually labeled data collected from actual test benches, and the other is simulated signal data obtained by constructing fault types such as excessively stiff or soft voltage regulating springs. The two data acquisition methods are used in a 1:2 ratio for model parameter training. When new data is acquired, it is processed according to the training parameters... Figure 3 The process involves predicting fault category labels for newly collected data.

[0055] This invention presents rule-based fault diagnosis and a feature fusion-based CNN diagnostic model. For the final fault type determination, a method of re-ranking the confidence of the diagnostic model is used to output the final result. (Reference) Figure 4 Each method outputs a confidence score for the type of fault based on its own model calculation rules. This invention reorders the confidence scores of different models and finally outputs the result with the highest confidence score as the final result.

Claims

1. A method for fault diagnosis of a hydraulic piston pump, characterized in that, Includes the following steps: S1. The plunger pump is tested according to the established standard test procedure, and then the key signals of the oil outlet, oil inlet, oil return, housing surface, and drive shaft are collected through the data acquisition system. S2. Extract frequency domain and time domain features from key signals; S3. Perform correlation analysis on different key signals and related components, and extract correlation features; S4. Confidence scores are assigned to frequency domain features, time domain features, and correlation features using two methods: a rule-based fault tree diagnostic model based on expert knowledge and a fault classification model that integrates expert knowledge and multi-source signal features, respectively. S5. Re-rank the confidence scores of the two different models and output the result with the highest confidence score as the final result. The key signals include the pressure, flow rate, and temperature of the oil outlet, oil inlet, and oil return port; the temperature of the housing surface; the vibration of the housing surface in the X, Y, and Z directions; and the torque and speed of the drive shaft. The time-domain features include dimensional time-domain features and dimensionless time-domain features; Dimensional time-domain characteristics include maximum value, maximum absolute value, minimum value, mean, peak-to-peak value, absolute mean, root mean square value, root mean square amplitude, and standard deviation. Dimensionless time-domain features include kurtosis, skewness, peak value, impulse, margin, and waveform. The frequency domain features include centroid frequency, average frequency domain amplitude, root mean square frequency, frequency variance, and maximum frequency component. The rule-based fault tree diagnostic model based on expert knowledge constructs a fault label system based on expert knowledge and experience, and sets a fault tree process for each label; After extracting the time-domain features, frequency-domain features, and correlation features of all signals, the fault tree process is used to analyze and diagnose each feature, and to make a fault type judgment and confidence score. The fault classification model, which integrates expert knowledge and multi-source signal features, uses a CNN convolutional model to perform multiple convolutional pooling operations on the acquired time-domain features, frequency-domain features, and correlation features to obtain feature vectors. Finally, the feature vectors are classified into fault category labels using softmax, and the final confidence score is output. The entire network model is trained with labeled data. The label data is obtained from two sources: one is by manually labeling data collected from actual test benches, and the other is by simulating fault types to obtain simulated signal data. The two data acquisition methods are used in a 1:2 ratio for training the model parameters.

Images