CN117628005A

CN117628005A - Signal-fused hydraulic motor fault diagnosis method and system

Info

Publication number: CN117628005A
Application number: CN202311646319.7A
Authority: CN
Inventors: 郭勇; 郭蒙宪; 彭延峰; 郭理宏; 杨来铭
Original assignee: Hunan University of Science and Technology
Current assignee: Hunan University of Science and Technology
Priority date: 2023-12-04
Filing date: 2023-12-04
Publication date: 2024-03-01

Abstract

The invention belongs to the technical field of fault diagnosis of a fusion signal hydraulic motor, and discloses a fault diagnosis method of the fusion signal hydraulic motor. According to the method for diagnosing the faults of the hydraulic motor with the fusion signal, 10 characteristic values such as root mean square, standard deviation and the like with the largest influence factors in the sound and vibration signals are selected by collecting the vibration and the sound signals of the hydraulic motor, and the fusion characteristic vector with higher and more obvious fault information content is constructed by adopting a similar label splicing method. And a fault diagnosis model of the plunger abrasion of the hydraulic motor is constructed by combining the fusion feature vector and the LightGBM algorithm, so that the rapid and accurate identification of the plunger abrasion of the hydraulic motor with 3 degrees is realized. Finally, the hydraulic motor plunger abrasion recognition results between different algorithms and different signals are compared and analyzed, and the superiority of the method is verified.

Description

Signal-fused hydraulic motor fault diagnosis method and system

Technical Field

The invention belongs to the technical field of fault diagnosis of a hydraulic motor with a fusion signal, and particularly relates to a fault diagnosis method of the hydraulic motor with the fusion signal.

Background

The hydraulic steering engine is a mechanical device commonly used in the fields of aerospace, submarines, ships and the like, and plays a vital role in steering and maintaining stability. The hydraulic motor is used as an executive component of the steering engine hydraulic system, and the working performance of the hydraulic motor determines the safe and stable operation of the whole steering engine system, so that the simple and effective fault diagnosis method is of great significance to the normal operation of the hydraulic system. The traditional hydraulic motor fault diagnosis method has strong dependence on experience knowledge and has the problem of difficult fault feature extraction. Researchers have combined machine learning to diagnose faults by collecting vibration signals or sound signals when the hydraulic motor is running. The vibration signal has higher signal-to-noise ratio and sensitivity, the sound sensor can collect in a non-contact way, the sound signal has higher bandwidth, the collected frequency range is wider, and the vibration signal is commonly used in a fault diagnosis method. Vibration signals of the hydraulic plunger pump are monitored by Zhu, jiang and the like, the vibration signals are decomposed and reconstructed by utilizing wavelet transformation, fault characteristics of time domain, frequency domain and the like are extracted, and classification of different faults of the hydraulic plunger pump is realized by combining a convolutional neural network. Wang and Liu obtain modal function components of the vibration signal by using empirical mode decomposition, further calculate entropy of the time-frequency matrix, and finally perform fault diagnosis on the centrifugal pump by using random forests. The sensitivity of the acceleration sensor is a fixed value, the early fault identification is not accurate enough, the vibration signal has strong nonlinearity and non-stationarity, and the signal acquisition of the sensor is greatly influenced. The sound sensor has the advantages of nondestructive installation and non-contact measurement, has higher sensitivity, and is applied to data acquisition of fault diagnosis by most students. Tang et al reviewed mechanical fault diagnosis based on audio signal analysis, analyzed the advantages that were possessed by using audio signals, and summarized the development prospects of audio signal-based fault diagnosis. Mian et al finally realize bearing fault diagnosis by utilizing a support vector machine algorithm by collecting sound signals of bearings under different fault states and extracting the tone quality characteristics of the signals. The sound signal contains many noise in the environment, and it is necessary to perform a decomposition noise reduction process in order to improve the signal-to-noise ratio of the sound signal. The number of decomposed layers is less, so that the denoising effect is difficult to achieve, and the number of decomposed layers is more, so that useful information in signals can be reduced, and the diagnosis accuracy is affected.

Because of the characteristics of uncertainty and confidentiality of the faults of the hydraulic system, the analysis of the fault points is complex, and the faults of the hydraulic system are difficult to accurately identify only through a single signal source or fault characteristics. In order to ensure the stable operation of the inner curve radial plunger type hydraulic motor and timely and accurately realize fault monitoring, a learner proposes a fault diagnosis method based on the fusion characteristics of the multisource sensors and achieves ideal effects. Li and the like are used for detecting the fault type of the sucker-rod pump system, a multi-feature fusion fault diagnosis model based on Fourier descriptors and graphic features of an indication graph is provided, the robustness of the features is enhanced, and the diagnosis accuracy of the multi-input feature fusion model is higher than that of a single-input feature model. Tang et al have utilized vibration signal, pressure signal and acoustic signal of the hydraulic plunger pump, combine Convolutional Neural Network (CNN) to propose a hydraulic plunger pump fault diagnosis method of self-adaptation learning rate. Three original signals are converted into two-dimensional time-frequency images through continuous wavelet transformation, and different fault types are identified by utilizing a CNN model. Karabacak et al [9] is to reduce the abrasion and failure risk of the worm gear box, and to process the vibration, sound and thermal image data of the normal and failure rotor, and to extract the time domain, frequency domain and thermal image features in singular, dual or triple form by using an Artificial Neural Network (ANN) and a Support Vector Machine (SVM), so as to realize the abrasion failure diagnosis of the worm gear box. Long et al propose a multi-sensor information driving motor fault diagnosis method based on AdaBoost. And extracting frequency domain characteristics of current, magnetic and vibration signals of the acquired motor by using Hilbert transformation and Fourier transformation, and training and diagnosing by using an AdaBoost model. The fault diagnosis method has high robustness and generalization capability. According to the analysis of the literature, the multi-signal fusion is obviously higher in diagnosis and identification accuracy than that of a single signal, but the processing of the original signal is complex, and the fault diagnosis method of the traditional machine learning such as a neural network and an AdaBoost algorithm is combined, so that the memory occupation is large, the operation time is long, and the fault identification rate is low.

The traditional GBDT and AdaBoost machine learning algorithms are realized on Boosting algorithms, and have the defects of poor robustness and low diagnosis efficiency. XGBoost and LightGBM algorithms are based on the GBDT algorithm, and the gradient lifting tree algorithm solves the problems of long training time and low fault recognition rate of the traditional algorithm and is the most widely used fault diagnosis method at present. The original data are processed by the Xiang and Wang respectively through the fast Fourier transform and the support vector machine algorithm, and then feature importance degree sequencing and diagnosis are combined with the XGBoost model, so that the accuracy rate of fault diagnosis of the rolling bearing is improved. Wu and Zhang, etc. are used for improving the accuracy and reliability of wind driven generator fault diagnosis, establishing an XGBoost fault identification model, and comparing the diagnosis result with a support vector machine and an Adaboost algorithm, and the result shows that the XGBoost algorithm has higher classification accuracy. The study of the scholars shows that the XGBoost algorithm has higher fault diagnosis accuracy than the traditional machine learning algorithm, and is widely applied to fault diagnosis of various mechanical equipment. However, the data traversing mode of the XGBoost algorithm limits the operation speed and reduces the efficiency of fault diagnosis. The LightGBM algorithm can improve the efficiency of data operation under the condition of ensuring the accuracy, and is widely applied to fault diagnosis of gearboxes, rolling bearings and hydraulic motors. Tang et al have studied the application of the LightGBM algorithm in wind power gear box fault diagnosis aiming at the problem that the efficiency and the precision of the traditional machine learning algorithm in wind power generator gear box fault diagnosis are low. It is further demonstrated that the LightGBM algorithm has higher detection accuracy than the conventional algorithm, and has lower false positive rate and false negative rate. Guo et al combined the wavelet noise reduction algorithm with the LightGBM model to diagnose the loosening fault of the hydraulic motor base bolt, and compared with the traditional machine learning algorithm, and the results show that the LightGBM algorithm has higher diagnosis efficiency. Xu and the like provide a rolling bearing fault diagnosis method combining a convolutional neural network and a LightGBM for solving the problems of long training time, low diagnosis efficiency and the like of a traditional fault diagnosis model based on a deep learning algorithm. And the diagnosis efficiency and accuracy of the LightGBM model are superior to those of other models through constructing a data set.

From the above document, it is known that the sensitivity of the vibration sensor is a fixed value, the early failure recognition is not accurate enough, and the vibration signal has strong nonlinearity and non-stationarity, which has a great influence on the sensor acquisition signal. The sound signal contains many noise in the environment, and it is necessary to perform a decomposition noise reduction process in order to improve the signal-to-noise ratio of the sound signal. The number of decomposed layers is less, so that the denoising effect is difficult to achieve, and the number of decomposed layers is more, so that useful information in signals can be reduced, and the diagnosis accuracy is affected. The single signal is adopted to carry out fault diagnosis and identification rate is not high, and the advantages of the two sensors can be combined by adopting a signal fusion method, and the fault identification rate is improved by utilizing different sensor sensitive frequency bands. For the existing multi-source sensor fusion method, the literature proves that the method has a higher diagnosis effect than a single signal characteristic model. However, the conversion and the feature extraction of the original data are troublesome, and the fault identification is mostly carried out based on the traditional GBDT and AdaBoost algorithm, so that the fault diagnosis model of the method has long training time and low diagnosis efficiency.

Through the above analysis, the problems and defects existing in the prior art are as follows:

The fault of the hydraulic motor is identified only through a single signal source or fault characteristics, and the fault identification result is poor in accuracy due to the uncertainty and the stealth characteristics of the fault of the hydraulic motor system;

the hydraulic motor fault identification is carried out by combining a single signal source with machine learning algorithms such as GBDT and AdaBoost, and the robustness and efficiency of the diagnosis process are low.

Disclosure of Invention

Aiming at the problems existing in the prior art, the invention provides a fault diagnosis method for a hydraulic motor with fusion signals.

The invention is realized in such a way that the fault diagnosis method of the hydraulic motor with the fusion signal comprises the following steps:

step one, simulating plunger faults through an inner curve radial plunger type hydraulic motor test bed, and collecting vibration signals and sound signals of a motor when the plunger is normal and three different degrees of abrasion are carried out;

step two, noise reduction processing is carried out by utilizing a wavelet soft threshold algorithm, and feature fusion is carried out by extracting time domain and frequency domain feature values according to data correlation;

step three, constructing a lightGBM motor plunger fault diagnosis model, dividing fusion characteristics into a data set containing normal and 3 fault signal labels, and training and diagnosing;

and step four, the superiority of the method is verified through comparison between different algorithms and different signals.

Further, the method for simulating plunger faults through the inner curve radial plunger type hydraulic motor test bed comprises the following steps:

the structure of the inner curve radial plunger hydraulic motor mainly comprises a stator, a rotor, an output shaft, a plunger and balls; the stator curve consists of 6 evenly distributed action arc sections, and 10 plungers are distributed around the rotor at equal intervals; when the motor works, high-pressure oil enters a rotor high-pressure cavity through the oil distribution shaft, larger pressure is generated to push the plunger to extend outwards along the radial direction of the shaft, so that the balls are contacted with the inner curve guide rail to generate interaction force, and tangential component force drives the rotor to rotate continuously to output torque; meanwhile, hydraulic oil in a plunger cavity in a low pressure area is discharged through a low pressure oil port of an oil distribution shaft, and the motor continuously works by continuously introducing high pressure oil and discharging low pressure oil;

vibration between the stator and the rotor is mainly caused by a plunger assembly, and as shown in a plunger motor structure diagram, 2 plungers are positioned at the same position, and typical vibration frequency is mainly as follows:

wherein n is the rotating speed of the hydraulic motor, p is the number of inner curves, and q is the number of plungers; the frequency for one of the plungers can be expressed as

Further, the method for extracting the time domain and frequency domain characteristic values according to the data correlation to perform characteristic fusion comprises the following steps:

(1) Feature selection

Performing data correlation visualization processing on the denoised signal in Python, and selecting 6 time domain features such as an average value, a root mean square, a peak value index, a standard deviation, a waveform index, a pulse index and the like and 6 frequency domain features such as an amplitude spectrum average value, an amplitude spectrum standard deviation, a frequency center of gravity, a spectral amplitude skewness, a frequency standard deviation, a frequency skewness and the like from the vibration signal and the sound signal data according to the feature importance;

(2) Model evaluation method

The confusion matrix is a common index for evaluating the performance of the model, can reflect the relation between a predicted result and a real situation, and is evaluated by adopting 4 indexes such as accuracy, precision, recall, f1-score and the like in the model evaluation;

the expression of each evaluation index is:

PR curve ROC curve is a technique and tool for evaluating classifier performance, which can calculate the accuracy and reliability of the predicted result; in PR curve, the recovery and precision are drawn on the same curve; ROC curves can intuitively compare the performance of a classifier by plotting the true rate (TPR) and False Positive Rate (FPR) of the classifier on a curve; TPR and FPR can be expressed as:

(3) Diagnostic method flow

Setting up a test bed and collecting signals: building an inner curve radial fault diagnosis test bed, carrying out three different-depth abrasion on the plunger, and collecting vibration and sound signals of the motor in different faults by using a vibration acceleration sensor and a PCB sound sensor;

Signal denoising and feature extraction: denoising the vibration signal and the sound signal by using a wavelet algorithm respectively, and extracting time domain and frequency domain characteristics by using a mathematical method;

signal feature fusion: fusing the characteristic values of the vibration signal and the sound signal by adopting a characteristic fusion method;

model training and fault diagnosis: and dividing the vibration signal data, the sound signal data and the fusion signal characteristic data set into a training set and a testing set respectively, inputting the training set and the testing set into a fault diagnosis model for training and diagnosis, and setting the maximum depth to prevent the occurrence of over fitting.

Further, the method for collecting vibration signals and sound signals of the motor when the plunger is worn normally and in three different degrees is as follows:

1) An experimental method;

2) Data processing;

3) And (5) selecting characteristics.

Further, the experimental method:

the invention builds an inner curve radial plunger hydraulic motor fault diagnosis experimental platform, which simulates fault types of different wear depths of plunger structures, and the experimental platform consists of a fault simulation system and a data acquisition system:

the plunger fault simulation test bed utilizes an inner curve radial plunger type hydraulic motor to simulate faults, a PCB sound sensor is arranged at the position 25cm outside the hydraulic motor, a vibration acceleration sensor is arranged on a rear end cover of the hydraulic motor in a magnetic attraction mode, an X axis is the same as the radial direction of the motor, and a rotating speed sensor is arranged at the position 4mm outside the axial direction of the hydraulic motor;

The PC end is connected with the data acquisition card through the Ethernet to receive and store data; the rotating speed display instrument can display the rotating speed of the hydraulic motor in real time, so that the rotating speed of each experiment is ensured to be the same, the interference of irrelevant factors is eliminated, and the accuracy of an experiment result is improved; during experiments, a servo control system is used for controlling a hydraulic oil source to supply stable flow, so that the rotating speed of a hydraulic motor is ensured to be stable at 300r/min; setting the sampling frequency of the data acquisition system to 2000Hz, and acquiring vibration and sound signals when the plunger is normal and the three different degrees of wear faults, wherein each group of experimental data is acquired for 6 minutes.

Further, the data processing method comprises the steps of:

experimental data obtained by the vibration sensor and the sound sensor usually contain interference of noise signals, and particularly, the detection frequency of the sound sensor is low, so that the experimental data is sensitive to sounds in the environment; therefore, according to the characteristics of the plunger abrasion fault signals, a wavelet soft threshold algorithm is selected to decompose and reconstruct vibration and sound signals so as to achieve the denoising effect on the original signals; the soft threshold function may be expressed as:

the signal after denoising through the wavelet soft threshold is a time domain signal, and for better observing the characteristics of the signal and extracting the frequency domain characteristics, a Fast Fourier Transform (FFT) is selected to transform the signal into a frequency domain representation, and the Fourier transform and the inverse transform formulas thereof are respectively as follows:

As can be seen from frequency domain comparison, burrs in the signals after the wavelet soft threshold denoising are reduced, so that the cleanliness of the signals is improved, and the denoising effect is achieved; the invention adopts an inner curve radial plunger type hydraulic motor, the number of plungers is 10, the inner curve of the shell is 6, and the experimental rotating speed is 300r/min; as can be seen from the comparison of fig. 6, the vibration sensor is sensitive to half frequency multiplication of the rotation speed of the hydraulic motor, the sound sensor is sensitive to one and two times frequency multiplication of the rotation speed of the hydraulic motor, and the frequency response effect of the single plunger is better; considering that the vibration sensor and the sound sensor have different response frequencies to the plunger fault, the invention combines the respective advantages of the vibration and the sound signals to fuse the characteristic values of the two signals, thereby improving the diagnosis accuracy of the plunger fault of the hydraulic motor.

Further, the feature selection method comprises the following steps:

simulating plunger faults by using an inner curve radial plunger hydraulic motor experiment table, and collecting vibration and sound signals of the plunger in different states; the acquisition frequency is set to 2000Hz, each group of data is acquired for 6 minutes, each group of data has 720000 sampling points, 4000 sampling points are sequentially selected to generate one sample, 180 samples are total, and the sample numbers of the training set and the test set are 80% and 20% respectively;

The time domain and frequency domain features are common indicators for reflecting the running state of the hydraulic motor; signals of normal and different abrasion of the plunger are collected by using vibration and sound sensors, after wavelet soft threshold denoising is performed, 11 types of time domain and frequency domain characteristic values of each group of data are respectively extracted by adopting a mathematical method; generating a vibration signal and sound signal characteristic value data set, and carrying out data set correlation analysis:

the data correlation thermodynamic diagram reflects the correlation information among the data, and the larger the positive correlation and the negative correlation values in the diagram are, the more the hydraulic motor running state information can be reflected; 10 vibration signal characteristic values such as a root mean square value, a root amplitude value, a standard deviation, a peak value index, a waveform index, a spectrum amplitude value average value, a spectrum amplitude value standard deviation, a spectrum amplitude value skewness, a spectrum amplitude value kurtosis, a frequency skewness and the like are selected according to the thermodynamic diagram; 10 kinds of sound signal characteristic values such as root mean square value, square root amplitude, peak value, standard deviation, peak value index, pulse index, spectrum amplitude mean value, spectrum amplitude standard deviation, frequency center of gravity, frequency standard deviation and the like are selected; and finally, carrying out transverse fusion on each group of normal and fault characteristic values according to the corresponding data labels, and only increasing the length of the data set without changing the number of samples.

Further, the method for training and diagnosing the fusion characteristics by dividing the fusion characteristics into a data set containing normal and 3 fault signal labels comprises the following steps:

(1) Pre-classifying and visualizing data;

(2) And (5) evaluating a model.

Further, the data pre-classification and visualization method comprises the following steps:

in order to observe the classification effect of the fusion signal data relative to vibration and sound signals in advance, the data is subjected to visual processing by using a T-SNE algorithm; as can be seen from the graph, the classification of the vibration signals has a certain aggregation, but the aggregation points are too many, and the samples in the same class are not aggregated together; the classification discreteness of the sound signals is strong, different types of data points are fused together, and the classification effect is poor; as can be seen from the classification of the fusion signal, the data points of different types are almost completely separated, the cross points between the samples are few, and the classification effect is better.

Further, the model evaluation method:

the AdaBoost algorithm is an improvement on the basis of the Boosting algorithm, and has higher effectiveness and practicability; the GBDT algorithm is a gradient lifting iterative decision tree algorithm, the XGBoost and the LightGBM algorithm are based on the improvement of GBDT, and the GBDT algorithm is the most popular decision tree lifting algorithm and has higher learning efficiency; the invention verifies the high efficiency of the LightGBM algorithm and the classification effect of the fusion signal data by comparing the classification accuracy of the fusion signal, the vibration signal and the sound signal respectively by the four algorithms;

The diagnosis results of different models are poor in the traditional AdaBoost algorithm, and the diagnosis accuracy of three data sets is lower than that of other three algorithms by more than 10%; the classification effect of the GBDT, XGBoost and LightGBM algorithms on the three data sets is not quite different, but the diagnosis accuracy of the LightGBM algorithm is 0.17% -0.33% higher, the diagnosis result accuracy of the three algorithms on the fused data sets is higher, the training time of the AdaBoost algorithm is shortest, and the diagnosis accuracy of the AdaBoost algorithm is lower; the diagnosis accuracy of the GBDT, XGBoost and the LightGBM algorithms is not much different, but the training time of the LightGBM algorithm is respectively shortened by 6 times and 3 times compared with the training time of the GBDT and the XGBoost algorithms, so that the LightGBM algorithm has higher diagnosis accuracy and higher efficiency;

denoising the vibration signal and the sound signal by using a wavelet soft threshold algorithm, extracting characteristic values, correspondingly fusing according to the data labels, and generating fused signal data sets, wherein each data set comprises normal and three fault types, but the number of samples is not changed; the failure diagnosis results of the LightGBM model on the fusion signal, the vibration signal and the sound signal are introduced into accuracy, precision, recall and f1-score as evaluation indexes of classification performance; training and diagnosing the three data sets by using the LightGBM model, and comparing the three data sets to see that the diagnosis accuracy of the fused data sets is improved by 4.86% and 14.59% compared with that of vibration and sound signals respectively; the diagnosis results of the fusion data set precision, recall and the f1-score are improved by 4.5-13.25% compared with the vibration data set and the sound data set;

ROC and PR curves are used to evaluate the classification performance of a machine learning algorithm on a given dataset, each dataset containing a fixed number of positive and negative samples, the larger the area under the curve, the better the classification performance of the machine learning algorithm on the dataset; FPR refers to the probability that an actual negative sample is predicted as a positive sample, TPR refers to the probability that an actual positive sample is predicted as a negative sample; precision refers to the proportion of samples predicted to be positive to positive samples, recall refers to the proportion of positive samples predicted to be positive; based on the classification effect of the LightGBM algorithm on the vibration, sound and fusion signal data sets, the classification effect of the LightGBM algorithm on the sound data sets is the worst, and the classification effect of the LightGBM algorithm on the fusion data sets is far higher than that of the vibration and sound data sets.

Another object of the present invention is to provide a hydraulic motor fault diagnosis system with a fused signal, comprising:

fault simulation component: the test bed comprises an inner curve radial plunger type hydraulic motor test bed, a test bed and a test bed, wherein the inner curve radial plunger type hydraulic motor test bed is used for simulating the normal state of a plunger and wear faults with different degrees, and collecting corresponding vibration and sound signals;

signal processing assembly module: noise reduction processing is carried out on the acquired signals by utilizing a wavelet soft threshold algorithm, so that the signal quality is improved for subsequent analysis;

And a feature extraction and fusion module: based on data correlation analysis, extracting characteristic values from the processed time domain and frequency domain signals, and carrying out characteristic fusion to form a comprehensive characteristic set for fault diagnosis;

fault diagnosis model module: constructing a motor plunger fault diagnosis model based on a LightGBM algorithm, and performing model training and diagnosis by using a fusion characteristic data set containing normal and three fault signal labels;

performance verification component module: tools for evaluating and verifying the performance of diagnostic models, such as confusion matrices, PR curves, and ROC curves, are included, as well as comparative analysis of different algorithms and signal types, to verify the effectiveness and superiority of the proposed method.

Further, the test bed comprises an inner curve radial plunger type hydraulic motor test bed which is used for simulating normal and different degrees of abrasion faults of the plunger, and is simultaneously provided with vibration and sound signal acquisition devices which are used for acquiring corresponding signals of the motor in various states.

Further, the system comprises a signal processing module, which utilizes a wavelet soft threshold algorithm to perform noise reduction processing on the acquired vibration and sound signals, and improves the signal quality so as to more accurately perform fault diagnosis.

Further, the device comprises a feature extraction and fusion module which is used for extracting time domain and frequency domain feature values from the vibration and sound signals after noise reduction based on data correlation analysis and carrying out feature fusion to form a comprehensive feature set for fault diagnosis.

Further, the motor plunger fault diagnosis model building module based on the LightGBM algorithm is used for training and diagnosing the fusion characteristic data set containing normal and three fault signal labels, so that accurate diagnosis of the hydraulic motor plunger fault is achieved.

Further, the diagnosis system comprises a model evaluation and verification module, wherein the evaluation and verification module evaluates and verifies the performance of the diagnosis model by using evaluation tools such as confusion matrix, PR curve and ROC curve, and the like, and comprises the calculation of indexes such as accuracy, precision, recall rate and f1-score, and the verification of the superiority of the diagnosis method by the comparison analysis of different algorithms and signal types.

In combination with the technical scheme and the technical problems to be solved, the technical scheme to be protected has the following advantages and positive effects:

first, the invention provides a method for diagnosing faults of a hydraulic motor with fusion signals. The plunger faults are simulated through an inner curve radial plunger type hydraulic motor test bed, vibration signals and sound signals of the motor when the plunger is normal and worn in three different degrees are collected, noise reduction processing is carried out by utilizing a wavelet soft threshold algorithm, feature selection is carried out according to data correlation and feature importance, and feature fusion is carried out by extracting feature values of a time domain and a frequency domain. And constructing a LightGBM motor plunger fault diagnosis model, dividing fusion characteristics into a data set containing normal and 3 fault signal labels for training and diagnosis, and finally verifying the superiority of the method through comparison between different algorithms and different signals. The fault diagnosis method of the fusion characteristic solves the problem of low accuracy of data acquisition diagnosis by using a single sensor, uses two different signals for diagnosis, can combine fault information of different sensors, and increases fault characteristics to improve the fault recognition rate. The fault diagnosis model is built by adopting the LightGBM, so that the occupied running memory is reduced, the training time of the model is prolonged, and the fault diagnosis efficiency is greatly improved on the premise of ensuring the classification precision.

The results show that: compared with AdaBoost, GBDT and XGBoost algorithm, the diagnosis accuracy of the LightGBM algorithm is improved by 0.17% -0.33%; the diagnosis accuracy of the fusion signal under the LightGBM model is improved by 4.86% and 14.59% compared with that of the vibration signal and the sound signal respectively.

Secondly, the invention provides a fault diagnosis method for the hydraulic motor with the fusion signal. The plunger faults are simulated through an inner curve radial plunger type hydraulic motor test bed, vibration signals and sound signals of the motor when the plunger is normal and worn in three different degrees are collected, noise reduction processing is carried out by utilizing a wavelet soft threshold algorithm, feature selection is carried out according to data correlation and feature importance, and feature fusion is carried out by extracting feature values of a time domain and a frequency domain. And constructing a LightGBM motor plunger fault diagnosis model, dividing fusion characteristics into a data set containing normal and 3 fault signal labels for training and diagnosis, and finally verifying the superiority of the method through comparison between different algorithms and different signals. The fault diagnosis method of the fusion characteristic solves the problem of low accuracy of data acquisition diagnosis by using a single sensor, uses two different signals for diagnosis, can combine fault information of different sensors, and increases fault characteristics to improve the fault recognition rate. The fault diagnosis model is built by adopting the LightGBM, so that the occupied running memory is reduced, the training time of the model is prolonged, and the fault diagnosis efficiency is greatly improved on the premise of ensuring the classification precision.

Thirdly, the expected benefits and commercial value after the technical scheme of the invention is converted are as follows: a) Due to the uncertainty and the stealth characteristics of the hydraulic motor system faults, the faults of the hydraulic motor are difficult to accurately identify only through a single signal source or fault characteristics. The fusion signal fault diagnosis method combines the advantages of the sound sensor and the vibration sensor, contains more fault information, and has simpler feature extraction and fusion and more reliable diagnosis result; b) 10 eigenvalues such as root mean square, standard deviation and the like with the largest influencing factors in the sound and vibration signals are selected, the similar label splicing is adopted for feature fusion to form a new feature vector, and the fault information content of the sample vector is higher and more obvious; c) The fusion sample vector is combined with the LightGBM algorithm model, so that multithread parallel calculation can be performed, the occupation of a computer memory is saved, the diagnosis time is greatly reduced, and the fault identification performance of the plunger motor is improved.

The technical scheme of the invention fills the technical blank in the domestic and foreign industries: a) The invention provides a diagnosis method for identifying the plunger wear fault of an inner curve plunger type hydraulic motor based on the characteristics of a sound-vibration fusion signal by a LightGBM algorithm, which opens up a feasible technical route for the fault diagnosis of the plunger type hydraulic motor; b) The fault diagnosis method based on machine learning is widely applied to rotary machines such as bearings, gears and the like, but has no application in the existing domestic and foreign literature for the diagnosis of plunger motors, particularly inner curve radial plunger hydraulic motors.

Fourth, the significant technical progress of the method for diagnosing faults of a hydraulic motor with fused signals provided by the invention can be summarized as follows:

1. accurate acquisition of fault simulation and experimental data

Authenticity of fault simulation: the plunger faults are simulated through the inner curve radial plunger type hydraulic motor test bed, so that the fault conditions in actual work can be accurately simulated, and test data which are closer to the actual conditions are obtained.

Multidimensional data acquisition: vibration signals and sound signals are collected, and two different signal types are combined, so that the accuracy and reliability of fault diagnosis are improved.

2. Efficient signal processing and feature fusion

Application of denoising algorithm: and the wavelet soft threshold algorithm is used for noise reduction treatment, so that noise in signals is effectively removed, and the quality of the signals is improved.

Innovation of feature fusion: and feature fusion is carried out by utilizing the time domain and frequency domain feature values, so that the diagnosis capability of the model is enhanced, and the comprehensiveness and depth of fault diagnosis are ensured.

3. Using advanced machine learning models

Application of the LightGBM model: and constructing a motor plunger fault diagnosis model based on the LightGBM, wherein the model has the characteristics of high efficiency and high precision, and is suitable for processing large-scale data.

Model training and optimization of diagnostics: the fusion characteristics are divided into data sets containing normal and 3 fault signal labels for training, so that the generalization capability and the diagnosis accuracy of the model are improved.

4. Comprehensive verification of system performance

And (3) comparing the comprehensive algorithm: by comparing different algorithms, the superiority of the method is verified, and the advancement and effectiveness of the method are ensured.

Comprehensive analysis of signal types: the comparative analysis between the different signals further verifies the importance of feature fusion in improving diagnostic accuracy.

The obvious technical progress of the hydraulic motor fault diagnosis method based on the fusion signal is mainly embodied in the aspects of improving the accuracy, reliability and efficiency of fault diagnosis, and the performance of hydraulic motor fault diagnosis is obviously improved by comprehensively utilizing various signals, an advanced signal processing method and a machine learning model.

Drawings

Fig. 1 is a flowchart of a fault diagnosis method for a hydraulic motor with a fused signal according to an embodiment of the present invention.

FIG. 2 is a block diagram of an inner curve radial plunger motor provided by an embodiment of the present invention.

Fig. 3 is a schematic diagram of a LightGBM algorithm according to an embodiment of the present invention.

Fig. 4 is a flowchart of a method provided by an embodiment of the present invention.

Fig. 5 is a diagram of experimental equipment provided by an embodiment of the present invention.

Fig. 6 is a graph of plunger wear provided by an embodiment of the present invention.

Fig. 7 is a comparison chart of signal denoising effects according to an embodiment of the present invention.

FIG. 8 is a data correlation thermodynamic diagram provided by an embodiment of the present invention.

FIG. 9 is a T-SNE cluster map provided by an embodiment of the invention. (a) vibration signal (b) sound signal (c) fusion signal.

Fig. 10 is a diagram of an confusion matrix provided by an embodiment of the present invention. (a) vibration signal (b) sound signal (c) fusion signal.

FIG. 11 is a graph of diagnostic results of different models provided by an embodiment of the present invention.

FIG. 12 is a graph of training time for different models provided by an embodiment of the present invention.

Fig. 13 is a LightGBM prediction accuracy map provided by an embodiment of the present invention.

Fig. 14 is a diagram showing the classification effect of the LightGBM algorithm on vibration, sound and fusion signal data sets according to the embodiment of the present invention. (a) ROC curve; (b) PR curve.

In fig. 2: 1. a ball; 2. a plunger; 3. a rotor; 4. an output shaft; 5. and a stator.

Detailed Description

The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

Based on the fusion signal hydraulic motor fault diagnosis method provided by the invention, the following two specific embodiments and implementation schemes thereof are as follows:

example 1: diagnosis of slight wear failure

1) And (3) fault simulation: on an inner curve radial plunger type hydraulic motor test bed, a slightly worn plunger fault is simulated.

2) And (3) data acquisition: vibration signals and sound signals in a normal state and in a slight abrasion state are collected by using a vibration acceleration sensor and a PCB sound sensor.

3) And (3) signal processing: the collected signals are noise-reduced using a wavelet soft threshold algorithm, and then feature values are extracted from the time and frequency domains.

4) Feature fusion and model construction: and fusing the time domain and frequency domain characteristic values, and constructing a motor plunger fault diagnosis model based on the LightGBM.

5) Model training and diagnosis: the model was trained and diagnosed using a dataset containing normal and slightly worn signal tags.

6) Performance evaluation: the accuracy and reliability of the model is assessed by confusion matrix and PR curve ROC curve.

Example 2: diagnosis of severe wear failure

1) And (3) fault simulation: severe wear plunger failure was simulated on the same test bench.

2) And (3) data acquisition: vibration and sound signals are also acquired in normal and heavy wear conditions.

3) And (3) signal processing: and carrying out the same noise reduction processing on the signals, and extracting corresponding time domain and frequency domain characteristic values.

4) Feature fusion and model optimization: and fusing the extracted features, and optimizing the LightGBM model to adapt to the fault features of heavy wear.

5) Model training and diagnosis: the model is trained and diagnosis is performed using a dataset containing heavily worn signal tags.

6) And (3) performance verification: and (3) performing performance verification on the model by using the model evaluation index, and ensuring that high accuracy is maintained under the condition of heavy wear.

The two embodiments provided by the invention show how to apply the method for diagnosing the faults of the hydraulic motor by fusion signals aiming at different degrees of abrasion faults (slight and severe), provide actual operation flow and steps and ensure high precision and effectiveness of fault diagnosis.

As shown in fig. 1, the present invention provides a method for diagnosing faults of a hydraulic motor with fused signals, comprising the following steps:

S101, simulating plunger faults through an inner curve radial plunger type hydraulic motor test bed, and collecting vibration signals and sound signals of a motor when the plunger is normal and three different degrees of abrasion are performed;

s102, performing noise reduction processing by using a wavelet soft threshold algorithm, and extracting time domain and frequency domain characteristic values according to data correlation to perform characteristic fusion;

s103, constructing a LightGBM motor plunger fault diagnosis model, dividing fusion characteristics into a data set containing normal and 3 fault signal labels, and training and diagnosing;

s104, comparing different algorithms and different signals to verify the superiority of the method.

The method for simulating plunger faults through the inner curve radial plunger type hydraulic motor test bed provided by the invention comprises the following steps:

the working principle of the inner curve radial plunger hydraulic motor is shown in figure 2, and the structure of the inner curve radial plunger hydraulic motor mainly comprises a stator, a rotor, an output shaft, a plunger and balls; the stator curve consists of 6 evenly distributed action arc sections, and 10 plungers are distributed around the rotor at equal intervals; when the motor works, high-pressure oil enters a rotor high-pressure cavity through the oil distribution shaft, larger pressure is generated to push the plunger to extend outwards along the radial direction of the shaft, so that the balls are contacted with the inner curve guide rail to generate interaction force, and tangential component force drives the rotor to rotate continuously to output torque; meanwhile, hydraulic oil in a plunger cavity in a low pressure area is discharged through a low pressure oil port of an oil distribution shaft, and the motor continuously works by continuously introducing high pressure oil and discharging low pressure oil;

/>

LightGBM algorithm principle

Both the XGBoost and the LightGBM algorithms are improvements based on the GBDT algorithm; the GBDT and XGBoost models both adopt a Level-wise growth strategy; the leaves of the same layer can be split at the same time by traversing the data once, so that the overfitting is reduced, and the optimization for a plurality of threads is easier; however, the data needs to be traversed layer by layer in the mode, so that a plurality of unnecessary searches or splits are generated, more running memory is consumed, the operation time is increased, and the efficiency is reduced; the LightGBM model is fused with a histogram algorithm, the principle is that the original data of the features are divided into k discrete features, k boxes are constructed for statistical features, the data do not need to be traversed during searching, the optimal splitting point can be found by traversing the k boxes, and the operation time is greatly shortened; the division mode of the LightGBM model adopts a Leaf-wise growth strategy; the method is that the Leaf with the maximum information gain is searched out from the current Leaf, and under the same splitting times, the Leaf-wise growth strategy has lower error than the Level-wise growth strategy, the occupied running memory is less, and the accuracy and the efficiency of the algorithm are improved; the disadvantage of the LightGBM model is that the phenomenon of over-fitting easily occurs, and therefore, the maximum depth parameter is added for limitation during model training;

The LightGBM model also incorporates two algorithms, gradient-based One-Side Sampling (GOSS) and mutually exclusive feature binding (Exclusive Feature Bundling, EFB), the principle of which is shown in fig. 3; GOSS is to sample the sample by using gradient information, and the influence of the sample with large gradient on the information gain is large; samples with large gradient are reserved during sampling, samples with small gradient are randomly sampled, the accuracy of the model is not lost, the memory occupation and the training time can be reduced, and the diagnosis efficiency of the model is greatly improved; the feature space in the data is often mutually exclusive, and the EFB algorithm binds a plurality of mutually exclusive features together from the aspect of reducing feature dimensions, so that the feature quantity is reduced, and the running speed of the model is improved.

The method for extracting the time domain and frequency domain characteristic values according to the data correlation and carrying out characteristic fusion comprises the following steps:

(1) Feature selection

Performing data correlation visualization processing on the denoised signal in Python, and selecting 6 time domain features such as an average value, a root mean square, a peak value index, a standard deviation, a waveform index, a pulse index and the like and 6 frequency domain features such as an amplitude spectrum average value, an amplitude spectrum standard deviation, a frequency center of gravity, a spectral amplitude skewness, a frequency standard deviation, a frequency skewness and the like from the vibration signal and the sound signal data according to the feature importance; the characteristics can reflect the working state of the inner curve plunger motor most, can reflect the failure information of the motor in time, and the mathematical formulas of the time domain and the frequency domain characteristics are shown in the table 1:

TABLE 1 time-domain and frequency-domain statistics

In the table, x (N) (n=1, 2, …, N) represents the original signal, N represents the number of data points; s (k) (k=1, 2,., M represents the signal spectral amplitude, M represents the number of spectral lines; f (k) represents the magnitude of the k-th spectral line frequency;

(2) Model evaluation method

The confusion matrix is a common index for evaluating the performance of the model, and as shown in table 2, can reflect the relation between the prediction result and the real situation, and 4 indexes such as accuracy, precision, recall, f1-score and the like are commonly selected for evaluating the model evaluation;

TABLE 2 confusion matrix

The expression of each evaluation index is:

(3) Diagnostic method flow

The diagnostic method flow is shown in fig. 4.

The method for collecting the vibration signals and the sound signals of the motor when the plunger is worn normally and in three different degrees is as follows:

1) An experimental method;

2) Data processing;

3) And (5) selecting characteristics.

The experimental method provided by the invention comprises the following steps:

the invention builds an inner curve radial plunger hydraulic motor fault diagnosis experiment platform, simulates fault types of different wear depths of plunger structures, and the experiment platform consists of a fault simulation system and a data acquisition system, wherein the installation positions of experiment equipment and sensors are shown in figure 5:

as shown in fig. 5 (a), the plunger fault simulation test bed simulates faults by using an inner curve radial plunger type hydraulic motor, a PCB sound sensor is arranged at the outer side 25cm of the hydraulic motor, a vibration acceleration sensor is arranged on a rear end cover of the hydraulic motor in a magnetic attraction mode, an X axis is the same as the radial direction of the motor, and a rotating speed sensor is arranged at the outer side 4mm of the axial direction of the hydraulic motor;

The data acquisition system is shown in fig. 5 (b), and the PC end is connected with the data acquisition card through the Ethernet to receive and store data; the rotating speed display instrument can display the rotating speed of the hydraulic motor in real time, so that the rotating speed of each experiment is ensured to be the same, the interference of irrelevant factors is eliminated, and the accuracy of an experiment result is improved; during experiments, a servo control system is used for controlling a hydraulic oil source to supply stable flow, so that the rotating speed of a hydraulic motor is ensured to be stable at 300r/min; setting the sampling frequency of a data acquisition system to 2000Hz, and acquiring vibration and sound signals when the plunger is normal and the three different degrees of wear faults, wherein each group of experimental data is acquired for 6 minutes; the model and parameters of each experimental apparatus are shown in table 3:

table 3 model of experimental apparatus and parameters thereof

Device name	Model number	Parameters (parameters)
			Vibration sensor	1C302	Frequency range: 20Hz-12KHz
Sound sensor	PCB378B02	Frequency range: 3.75Hz-20KHz
			Plunger hydraulic motor	1001-0.1	Maximum rotational speed: 360r/min
Data acquisition card	FK2012	16 channels
			Rotation speed sensor	ZSM12-CS-01	Output current: 50mA
Rotating speed display	WR5135-FR-N	Measurement range: 0.2Hz-100KHz

In order to make the experimental effect better, a transverse abrasion mode is adopted to manufacture faults along the axial direction of the plunger; plunger failures are classified into mild wear, moderate wear and deep wear according to the depth of wear, with specific wear effects shown in fig. 6.

The data processing method provided by the invention comprises the following steps:

FIG. 7 is a graph comparing denoising effects of vibration and sound signals, respectively; as can be seen from frequency domain comparison, burrs in the signals after the wavelet soft threshold denoising are reduced, so that the cleanliness of the signals is improved, and the denoising effect is achieved; the invention adopts an inner curve radial plunger type hydraulic motor, the number of plungers is 10, the inner curve of the shell is 6, and the experimental rotating speed is 300r/min; as can be seen from the comparison of fig. 7, the vibration sensor is sensitive to half frequency multiplication of the rotation speed of the hydraulic motor, the sound sensor is sensitive to one and two times frequency multiplication of the rotation speed of the hydraulic motor, and the frequency response effect of the single plunger is better; considering that the vibration sensor and the sound sensor have different response frequencies to the plunger fault, the invention combines the respective advantages of the vibration and the sound signals to fuse the characteristic values of the two signals, thereby improving the diagnosis accuracy of the plunger fault of the hydraulic motor.

The feature selection method provided by the invention comprises the following steps:

simulating plunger faults by using an inner curve radial plunger hydraulic motor experiment table, and collecting vibration and sound signals of the plunger in different states; the acquisition frequency is set to 2000Hz, each group of data is acquired for 6 minutes, each group of data has 720000 sampling points, 4000 sampling points are sequentially selected to generate one sample, 180 samples are total, and the sample numbers of the training set and the test set are 80% and 20% respectively; the plunger type and sample number are shown in table 4:

table 4 dataset information

Status of	Training set/test set	Label (Label)
			Normal state	144/36	0
Failure A	144/36	1
			Failure B	144/36	2
Failure C	144/36	3

The time domain and frequency domain features are common indicators for reflecting the running state of the hydraulic motor; signals of normal and different abrasion of the plunger are collected by using vibration and sound sensors, after wavelet soft threshold denoising is performed, 11 types of time domain and frequency domain characteristic values of each group of data are respectively extracted by adopting a mathematical method; generating a vibration signal and sound signal characteristic value data set, and carrying out data set correlation analysis; the information-dependent thermodynamic diagram between data sets is shown in fig. 8:

the data correlation thermodynamic diagram of fig. 8 reflects the correlation information between the data, wherein the larger the positive correlation and the negative correlation values are, the more the hydraulic motor operation state information can be reflected; 10 vibration signal characteristic values such as a root mean square value, a root amplitude value, a standard deviation, a peak value index, a waveform index, a spectrum amplitude value average value, a spectrum amplitude value standard deviation, a spectrum amplitude value skewness, a spectrum amplitude value kurtosis, a frequency skewness and the like are selected according to the thermodynamic diagram; 10 kinds of sound signal characteristic values such as root mean square value, square root amplitude, peak value, standard deviation, peak value index, pulse index, spectrum amplitude mean value, spectrum amplitude standard deviation, frequency center of gravity, frequency standard deviation and the like are selected; finally, transversely splicing each group of normal and fault characteristic values according to the corresponding data labels to form a new characteristic vector without changing the number of samples;

The invention provides a method for training and diagnosing fusion characteristics by dividing the fusion characteristics into a data set containing normal and 3 fault signal labels, which comprises the following steps:

(1) Pre-classifying and visualizing data;

(2) And (5) evaluating a model.

The data pre-classification and visualization method provided by the invention comprises the following steps:

in order to observe the classification effect of the fusion signal data relative to vibration and sound signals in advance, the data is subjected to visual processing by using a T-SNE algorithm, and a T-SNE clustering diagram is shown in FIG. 9; as can be seen from the graph, the classification of the vibration signals has a certain aggregation, but the aggregation points are too many, and the samples in the same class are not aggregated together; the classification discreteness of the sound signals is strong, different types of data points are fused together, and the classification effect is poor; as can be seen from the classification of the fusion signals, the data points of different types are almost completely separated, the cross points between samples are few, and the classification effect is better;

FIG. 10 is a four-classification confusion matrix for vibration signal, sound signal, and fusion signal datasets for visualizing the classification effect of three datasets; sample numbers for test set labels 0-3 are 28, 32, 44 and 40, respectively; as can be seen from the vibration signal confusion matrix of fig. 10 (a), 3 faults are predicted as actual normal, and 9 faults are classified as error; as can be seen from the acoustic signal confusion matrix of fig. 10 (b), all samples that are actually normal are predicted correctly, but there are 23 classification errors among the three faults; as can be seen from the fusion signal confusion matrix of fig. 10 (C), all samples that are actually normal are predicted correctly, and there are only 5 classification errors between the faults B and C among the three faults; the classification accuracy of the fused signal dataset is far higher than that of the vibration signal and the sound signal.

The model evaluation method provided by the invention comprises the following steps:

as shown in FIG. 11, the diagnosis results of different models are poor in the traditional AdaBoost algorithm, and the diagnosis accuracy of the three data sets is lower than that of the other three algorithms by more than 10%; the classification effect of the GBDT, XGBoost and LightGBM algorithms on the three data sets is not great, but the diagnosis accuracy of the LightGBM algorithm is 0.17% -0.33% higher, and the diagnosis result accuracy of the three algorithms on the fusion data sets is higher; as can be seen from the different model training durations of fig. 12, the AdaBoost algorithm has the shortest training time, but its diagnostic accuracy is lower; the diagnosis accuracy of the GBDT, XGBoost and the LightGBM algorithms is not much different, but the training time of the LightGBM algorithm is respectively shortened by 6 times and 3 times compared with the training time of the GBDT and the XGBoost algorithms, so that the LightGBM algorithm has higher diagnosis accuracy and higher efficiency;

Denoising the vibration signal and the sound signal by using a wavelet soft threshold algorithm, extracting characteristic values, correspondingly fusing according to the data labels, and generating fused signal data sets, wherein each data set comprises normal and three fault types, but the number of samples is not changed; FIG. 13 is a graph showing the results of fault diagnosis of the LightGBM model on the fusion signal, the vibration signal and the sound signal, and introducing accuracy, precision, recall and f1-score as evaluation indexes of classification performance; training and diagnosing the three data sets by using the LightGBM model, and comparing the three data sets to see that the diagnosis accuracy of the fused data sets is improved by 4.86% and 14.59% compared with that of vibration and sound signals respectively; the diagnosis results of the fusion data set precision, recall and the f1-score are improved by 4.5-13.25% compared with the vibration data set and the sound data set;

ROC and PR curves are used to evaluate the classification performance of a machine learning algorithm on a given dataset, each dataset containing a fixed number of positive and negative samples, the larger the area under the curve, the better the classification performance of the machine learning algorithm on the dataset; FPR refers to the probability that an actual negative sample is predicted as a positive sample, TPR refers to the probability that an actual positive sample is predicted as a negative sample; precision refers to the proportion of samples predicted to be positive to positive samples, recall refers to the proportion of positive samples predicted to be positive; the effect of classification of vibration, sound and fusion signal datasets based on the LightGBM algorithm is shown in fig. 14; as can be seen from the figure, the LightGBM algorithm has the worst classification effect on the sound dataset, and the classification effect on the fusion dataset is much higher than the vibration and sound dataset.

The invention provides a fault diagnosis method for a hydraulic motor with a fused signal. The plunger faults are simulated through an inner curve radial plunger type hydraulic motor test bed, vibration signals and sound signals of the motor when the plunger is normal and worn in three different degrees are collected, noise reduction processing is carried out by utilizing a wavelet soft threshold algorithm, feature selection is carried out according to data correlation and feature importance, and feature fusion is carried out by extracting feature values of a time domain and a frequency domain. And constructing a LightGBM motor plunger fault diagnosis model, dividing fusion characteristics into a data set containing normal and 3 fault signal labels for training and diagnosis, and finally verifying the superiority of the method through comparison between different algorithms and different signals. The fault diagnosis method of the fusion characteristic solves the problem of low accuracy of data acquisition diagnosis by using a single sensor, uses two different signals for diagnosis, can combine fault information of different sensors, and increases fault characteristics to improve the fault recognition rate. The fault diagnosis model is built by adopting the LightGBM, so that the occupied running memory is reduced, the training time of the model is prolonged, and the fault diagnosis efficiency is greatly improved on the premise of ensuring the classification precision.

The invention provides a fault diagnosis system of a hydraulic motor with fused signals, which comprises an inner curve radial plunger type hydraulic motor test bed, wherein the inner curve radial plunger type hydraulic motor test bed is used for simulating normal and different degrees of abrasion faults of a plunger, and is simultaneously provided with a vibration and sound signal acquisition device for acquiring corresponding signals of the motor in various states. The system comprises a signal processing module, wherein the collected vibration and sound signals are subjected to noise reduction processing by utilizing a wavelet soft threshold algorithm, and the signal quality is improved so as to more accurately carry out fault diagnosis. The system comprises a feature extraction and fusion module, a data correlation analysis module and a fault diagnosis module, wherein the feature extraction and fusion module is used for extracting time domain and frequency domain feature values from noise-reduced vibration and sound signals and carrying out feature fusion to form a comprehensive feature set for fault diagnosis. The motor plunger fault diagnosis model building module based on the LightGBM algorithm is used for training and diagnosing fusion characteristic data sets containing normal and three fault signal labels, so that accurate diagnosis of hydraulic motor plunger faults is achieved. The diagnosis model performance evaluation and verification module utilizes evaluation tools such as confusion matrix, PR curve, ROC curve and the like to evaluate and verify the performance of the diagnosis model, and comprises the calculation of indexes such as accuracy, precision, recall rate, f1-score and the like, and the verification of the superiority of the diagnosis method through the comparison analysis of different algorithms and signal types.

The connection relation and the working principle of the hydraulic motor fault diagnosis system with the fusion signal provided by the invention are as follows:

1) Fault simulation component: this assembly contains an inner curve radial plunger hydraulic motor test stand for simulating the normal condition of the plunger and varying degrees of wear failure. It can collect the corresponding vibration and sound signals of motor in various states.

2) Signal processing assembly module: this module receives vibration and sound signals from the faulty analog component. Then, it uses wavelet soft threshold algorithm to make noise reduction treatment to these signals so as to raise signal quality and make them more suitable for subsequent fault diagnosis.

3) And a feature extraction and fusion module: this module receives the processed signal from the signal processing component module. Based on the data correlation analysis, extracting characteristic values from the processed time domain and frequency domain signals, and carrying out characteristic fusion to form a comprehensive characteristic set for fault diagnosis.

4) Fault diagnosis model module: this module receives the comprehensive feature set from the feature extraction and fusion module. Then, it builds a motor plunger fault diagnosis model based on the LightGBM algorithm, and uses the fused feature data set containing normal and three fault signal tags for model training and diagnosis.

5) Performance verification component module: this module is used to evaluate and verify the performance of the diagnostic model. The accuracy, precision, recall and f1-score index of the diagnostic model are calculated by using tools such as confusion matrix, PR curve and ROC curve. In addition, it also verifies the superiority of the diagnostic method by comparing analysis of different algorithms and signal types.

The working flow of the whole system is as follows: first, the fault simulation assembly simulates and collects vibration and sound signals of the hydraulic motor. The signal processing component module then performs noise reduction processing on these signals. Then, the feature extraction and fusion module extracts and fuses features from the processed signals. The fault diagnosis model module then uses these features to train and diagnose the model. Finally, the performance verification component module evaluates and verifies the performance of the diagnostic model and verifies the superiority of the diagnostic method. It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.

In order to solve the problem of low accuracy of diagnosing faults by utilizing single signal data, improve the diagnosis efficiency of plunger wear faults of the inner curve radial plunger type hydraulic motor and ensure the stable operation of a hydraulic system, the hydraulic motor plunger wear fault diagnosis method based on the characteristics of identifying vibration and sound signal fusion of the lightGBM is provided; the method comprises the steps of performing feature selection according to data correlation and feature importance by collecting vibration signals and sound signals of a motor when a plunger of a fault diagnosis test bed is worn normally and in three different degrees, extracting time domain and frequency domain feature values, and performing feature fusion; constructing a LightGBM motor plunger fault diagnosis model, dividing fusion characteristics into data sets containing normal and 3 fault signal labels for training and diagnosis, and finally verifying the superiority of the method through comparison between different algorithms and different data sets; the specific conclusions are as follows:

setting up an inner curve radial plunger type hydraulic motor plunger fault test bed, and collecting vibration signals and sound signals of the plunger in normal and three different degree abrasion states when the motor is in normal operation;

carrying out noise reduction treatment on the two signals by utilizing a wavelet algorithm, extracting time domain and frequency domain characteristic values of the signals, and carrying out same-class data fusion to generate a fusion data set;

The method has the advantages that a LightGBM fault diagnosis model is built and compared with AdaBoost, GBDT and XGBoost models, the result shows that the fault diagnosis accuracy of the LightGBM model is 0.17% -0.33% higher, the training time is shortened by 3-5 times, and the LightGBM model is verified to have higher diagnosis efficiency;

the comparison shows that the diagnosis accuracy of the LightGBM model on the fusion signal data set is improved by 4.86 percent and 14.59 percent compared with that of the vibration signal and the sound signal respectively, and the evaluation indexes of precision, recall, f1-score and the like are also improved by 4.5 to 13.25 percent compared with that of the vibration and sound data set;

the ROC and PR curve analysis shows that the classification effect of the light GBM algorithm on the fusion data set is better than that of vibration and sound data;

the method provided by the invention achieves expected effects in fault diagnosis of the plunger type hydraulic motor, but has some defects, and future researches should consider the following problems: 1) When the sound signal is collected, the sound sensor is directly exposed to the environment, and a sound insulation cover can be added to reduce the interference of environmental noise; 2) Each group of signals in the experiment is collected for 6 minutes, the number of data points is not more, and the number of samples can be increased to improve the fault diagnosis accuracy.

The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.

Claims

1. The fault diagnosis method of the hydraulic motor with the integrated signal is characterized in that firstly, plunger faults are simulated through an inner curve radial plunger type hydraulic motor test bed, and vibration and sound signals of the motor under normal states and different wear degrees are collected; secondly, carrying out noise reduction treatment on the signals by utilizing a wavelet soft threshold algorithm, extracting characteristic values from a time domain and a frequency domain based on data correlation analysis, and carrying out characteristic fusion; then, constructing a motor plunger fault diagnosis model based on the LightGBM, and performing model training and diagnosis by using a data set containing normal and three fault signal labels; finally, the effectiveness and superiority of the method are verified through comparison analysis of different algorithms and different signal types.

2. The method of claim 1, comprising an inner curve radial plunger hydraulic motor test bed for simulating plunger failure including normal conditions and different degrees of wear, and collecting associated vibration and sound signals to provide raw data for subsequent failure diagnosis.

3. The method for diagnosing faults of a hydraulic motor with fused signals as claimed in claim 1, wherein the collected vibration and sound signals are subjected to noise reduction treatment through a wavelet soft threshold algorithm, so that noise interference in the signals is effectively removed.

4. The method of claim 1, wherein the time domain and frequency domain eigenvalues are extracted from the processed vibration and sound signals based on data correlation analysis, and the eigenvalues are fused to form a comprehensive feature set for more accurate fault diagnosis.

5. A method for diagnosing faults in a hydraulic motor with fused signals as claimed in claim 3, wherein a motor plunger fault diagnosis model based on the LightGBM algorithm is constructed, model training and diagnosis are performed using data sets containing normal and three fault signal labels, and the effectiveness and superiority of the method are verified by comparison analysis of different algorithms and different signal types.

6. A fused signal hydraulic motor fault diagnosis system based on the method of claim 1, comprising:

performance verification component module: tools for evaluating and verifying the performance of diagnostic models, including confusion matrices, PR curves, and ROC curves, are included, as well as comparative analysis of different algorithms and signal types, to verify the effectiveness and superiority of the proposed method.

7. A fused signal hydraulic motor fault diagnosis system based on the method of claim 1, comprising:

an inner curve radial plunger type hydraulic motor test bed is used for simulating normal and different degrees of abrasion faults of a plunger, and is simultaneously provided with a vibration and sound signal acquisition device for acquiring corresponding signals of the motor in various states;

The signal processing module is used for carrying out noise reduction processing on the collected vibration and sound signals by utilizing a wavelet soft threshold algorithm, so that the signal quality is improved, and fault diagnosis is more accurately carried out;

the feature extraction and fusion module is used for extracting time domain and frequency domain feature values from the noise-reduced vibration and sound signals based on data correlation analysis, and carrying out feature fusion to form a comprehensive feature set for fault diagnosis;

the motor plunger fault diagnosis model building module is used for training and diagnosing fusion characteristic data sets containing normal and three fault signal labels, so that accurate diagnosis of hydraulic motor plunger faults is realized;

and a model evaluation and verification module for evaluating and verifying the performance of the diagnosis model by using confusion matrix, PR curve and ROC curve evaluation tools, including calculation of accuracy, precision, recall rate and f-score index, and verifying the superiority of the diagnosis method by comparison analysis of different algorithms and signal types.