Motor System Fault Qualitative Diagnosis Method Based on Classification Analysis of Current Sample Characteristics

Abstract

The present invention discloses a method for qualitative diagnosis of motor system faults based on classification analysis of current sample features, comprising the following steps: S1, performing multi-point sampling on the current samples of the motor system to obtain the original current signal and converting it into a comprehensive current signal; S2, constructing a pre-training data set and a deep learning model to obtain the pre-training weight coefficient matrix; S3, classifying the fault samples according to the fault type, fault degree and fault location information; S4, establishing a fault current vector and a typical fault average current vector; S5, sampling the motor current in real time, calculating the fault comprehensive current signal and constructing a fault prediction current vector; S6, calculating the vector difference between the fault prediction current vector and the typical fault average current vector, and determining the fault type through the minimum value of the vector difference. The present invention can achieve rapid identification and accurate diagnosis of motor faults, has the advantages of rapid response and strong adaptability, and provides efficient support for the fault diagnosis and maintenance of the motor system.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical signal data processing (G06F), and particularly relates to a method for qualitatively diagnosing motor system faults based on classification analysis of current sample features. Background Art

[0002] As a key driving device in modern industry, motors play an indispensable role in various mechanical systems. However, during long-term operation, motors will inevitably encounter some faults. Understanding these common faults and their causes is of great significance for preventing fault occurrence, timely maintenance, and extending the service life of motors. Motor faults include a series of faults such as motor overheating, motor vibration, difficult motor starting, and excessive motor noise.

[0003] Motor condition monitoring and fault diagnosis technology is a technology for understanding and mastering the condition of motors during use, determining whether the whole or part of them is normal or abnormal, early detecting faults and their causes, and being able to predict the development trend of faults. Motor condition monitoring and fault diagnosis technology includes two aspects: identifying motor condition monitoring and predicting development trends. In motor fault diagnosis, common methods include signal processing-based methods and machine learning-based methods. Signal processing-based methods require complex preprocessing and analysis of signals and may not be able to accurately diagnose complex fault types. Machine learning-based methods, on the other hand, require a large amount of sample data for training, which not only increases the load on the system but may also result in misdiagnosis for some rare fault types. Therefore, a comprehensive and efficient diagnosis method is needed to qualitatively diagnose and diagnose motor systems. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention uses classification analysis based on current sample features to qualitatively diagnose and diagnose motor systems, which can also be detected and diagnosed during the working state of the motor, is very convenient, and has a low cost.

[0005] To achieve the above object, the present invention provides the following technical solution: A method for qualitatively diagnosing motor system faults based on classification analysis of current sample features, including the following steps:

[0006] S1. Multiply-point sample the current samples during the operation of the motor system through multiple sensors to obtain the original current signal, and convert the original current signal into a comprehensive current signal; time-share sample the motor current when the motor fails to obtain the fault current signal;

[0007] S2. Construct the comprehensive current signal and the fault current signal into a pre-training data set;

[0008] S3. Establish a deep learning model based on a convolutional neural network, import the processed pre-training data set into the model for training to obtain a pre-training weight coefficient matrix, and complete the construction of the pre-training model;

[0009] S4. Qualitatively classify the fault samples based on the fault type, fault degree, and fault location information;

[0010] S5. Establish a fault current vector for the fault current signal of the fault samples after qualitative classification, and obtain the typical fault average current vector;

[0011] S6. Perform real-time sampling on the current of the motor system, obtain the fault sampling current when an unknown fault occurs, calculate the fault comprehensive current signal, use the fault comprehensive current signal as a sample, calculate the fault prediction current using the pre-training model, and construct a fault prediction current vector;

[0012] S7. Calculate the vector difference between the fault prediction current vector and any one of the typical fault average current vectors, select the typical fault with the smallest vector difference as the determined fault, and complete the qualitative fault diagnosis for the motor system.

[0013] Further, the current signal acquisition in step S1 is specifically as follows:

[0014] S11. Connect the motor driver and connect the current sensor to the power line;

[0015] S12. Acquire the current signal, and use the sampling circuit and sampling chip in the driver to sample the connected current sensor to obtain the original current signal;

[0016] S13. Convert the acquired original current signal into a comprehensive current signal, and the conversion formula satisfies:

[0017] (1)

[0018] where, represents the discrete index of the comprehensive current signal, which is used to refer to the output sampling point calculated by the conversion formula, represents the value of the comprehensive current signal corresponding to the discrete index ; represents the sampling index of the original signal, represents the value of the original current signal corresponding to the sampling index ; is the imaginary symbol, is the natural base;

[0019] S14. Perform time-sharing sampling on the motor current when the motor fails to obtain the fault current signal , which satisfies:

[0020]

[0021] Among them, respectively represent the first to the (N - 1)-th sampling moments.

[0022] Furthermore, in step S2, the pre-training samples are constructed to satisfy:

[0023] ;

[0024] Among them, to respectively represent the fault current signals collected at the first to the (N - 1)-th sampling moments when the motor fails; represents the weight coefficient matrix, which is a row column matrix, represents the weight coefficient of the -th row and the -th column, and so on.

[0025] Furthermore, in step S5, the fault current vector satisfies:

[0026] ;

[0027] Among them, represents the fault current vector of the -th sample of the -th typical fault, represents the sampling current at the -th sample of the -th typical fault at the -th moment, and so on;

[0028] The typical fault average current vector satisfies:

[0029] ;

[0030] Among them, represents the typical fault average current vector of the -th typical fault, represents the total number of samples of the -th typical fault.

[0031] Furthermore, in step S6, the calculated fault comprehensive current signal satisfies:

[0032] ;

[0033] Among them, represents the value of the original current signal at the sampling index when an unknown fault occurs, Discrete index indicating the occurrence of an unknown fault Value of the comprehensive current signal

[0034] Furthermore, in step S6, using the fault comprehensive current signal as a sample, the calculated fault prediction current using the pre-trained model satisfies:

[0035] ;

[0036] Wherein, to respectively represent the fault prediction currents collected at the first to the N-1th sampling moments obtained by using the pre-trained model when an unknown fault occurs in the motor; represents the pre-trained weight coefficient matrix, which is a row column matrix, represents the pre-trained weight coefficient of the row and the column, and so on.

[0037] Furthermore, in step S6, constructing the fault prediction current vector satisfies:

[0038] ;

[0039] Wherein, represents the fault prediction current vector when an unknown fault occurs in the motor.

[0040] Furthermore, calculating the vector difference between the fault prediction current vector and the average current vector of any typical fault satisfies:

[0041] ;

[0042] Wherein, represents the vector difference between the fault prediction current vector when an unknown fault occurs in the motor and the average current vector of the th typical fault.

[0043] Furthermore, in step S7, selecting a typical fault with the smallest vector difference as the determined fault satisfies:

[0044] ;

[0045] Wherein, represents the fault qualitative diagnosis result of the unknown fault of the motor; represents the fault type that makes the vector difference value between the fault prediction current vector and the average current vector of the typical fault achieve the minimum value.

[0046] Beneficial effects

[0047] Compared with the prior art, the present invention provides a method for qualitative diagnosis of motor system faults based on classification analysis of current sample features, which has the following beneficial effects:

[0048] 1. Advantages of multi-point sampling and integrated signal conversion: Multiple sensors are used to perform multi-point sampling on the motor system and convert the original current signal into an integrated current signal, making the collected data more comprehensive and accurate. This step ensures that the current signals collected at different positions can be effectively fused, providing higher signal quality and helping to accurately identify the fault state of the motor.

[0049] 2. Improving diagnostic accuracy with pre-trained deep learning models: A deep learning model based on a convolutional neural network is pre-trained, enabling the model to efficiently identify different types of motor faults, and enhancing the model's generalization ability for motor fault patterns through the learning process of the pre-trained weight coefficient matrix. This can significantly improve the accuracy and reliability of the model during actual diagnosis.

[0050] 3. Efficiency of matching fault current vectors with typical faults: In this solution, by calculating the vector difference between the fault current vector and the average current vector of typical faults, the fault type of the motor can be quickly identified. By matching the typical fault with the smallest vector difference, the speed and accuracy of qualitative fault diagnosis are improved, providing an effective basis for timely maintenance.

[0051] 4. Real-time monitoring and fault prediction capabilities: The real-time sampling and calculation of the fault integrated current signal in the solution can instantaneously capture the fault signal when a fault occurs and perform fault prediction through the pre-trained model. This real-time response mechanism makes the system more sensitive and effectively improves the ability to quickly respond to and predict unknown faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to better understand the purpose, structure and function of the present invention, the method for qualitative diagnosis of motor system faults based on classification analysis of current sample features of the present invention will be further described below in conjunction with the drawings and specific embodiments.

[0054] Embodiment 1

[0055] Refer to Figure 1 , the present invention provides a method for qualitative diagnosis of motor system faults based on classification analysis of current sample features, including the following steps:

[0056] S1. Perform multi-point sampling of the current samples of the motor system during operation through multiple sensors to obtain the original current signal, and convert the original current signal into a comprehensive current signal; perform time-sharing sampling of the motor current when the motor fails to obtain a fault current signal;

[0057] S2, constructing the comprehensive current signal and the fault current signal into a pre-training data set;

[0058] S3. Establish a deep learning model based on a convolutional neural network, import the processed pre-training data set into the model for training, obtain a pre-training weight coefficient matrix, and complete the construction of the pre-training model;

[0059] S4, performing qualitative fault classification on the fault samples based on the fault type, fault degree and fault location information;

[0060] S5, establishing a fault current vector for the fault current signal of the fault sample after the fault qualitative classification, and obtaining a typical fault average current vector;

[0061] S6. Sample the current of the motor system in real time, obtain the fault sampling current when an unknown fault occurs, calculate the fault comprehensive current signal, use the fault comprehensive current signal as a sample, use the pre-trained model to calculate the fault prediction current, and construct the fault prediction current vector;

[0062] S7. Calculate the vector difference between the fault prediction current vector and any typical fault average current vector, select a typical fault with the smallest vector difference as the judgment fault, and complete the qualitative fault diagnosis for the motor system.

[0063] The current signal acquisition in step S1 is specifically as follows:

[0064] S11, connect the motor driver and connect the current sensor to the power line;

[0065] S12, collecting current signals, using the sampling circuit and sampling chip in the driver to sample the current signals of the connected current sensors to obtain the original current signals;

[0066] S13, converting the collected original current signal into a comprehensive current signal, and the conversion formula satisfies:

[0067] ; (1)

[0068] in, Represents the discrete index of the integrated current signal, which is used to refer to the output sampling point calculated by the conversion formula. Represents a discrete index The value of the integrated current signal; represents the sampling index of the original signal, Indicates the sampling index Value of the original current signal Is the imaginary symbol Is the natural base

[0069] S14. Perform time-division sampling on the motor current when the motor fails to obtain a fault current signal , satisfying:

[0070]

[0071] Among them, Respectively represent the first to the N-1th sampling moments

[0072] In step S2, constructing the pre-training samples satisfies:

[0073] ;

[0074] Among them, To Respectively represent the fault current signals collected at the first to the N-1th sampling moments when the motor fails Represents the weight coefficient matrix, which is Row Column matrix, Represents the Row and the Column weight coefficient, and so on

[0075] In step S5, the fault current vector satisfies:

[0076] ;

[0077] Among them, Represents the fault current vector of the th typical fault and the th sample, Represents the th typical fault and the th sample at the moment sampling current, and so on;

[0078] The typical fault average current vector satisfies:

[0079] ;

[0080] Among them, Represents the typical fault average current vector of the th typical fault, Represents the total number of samples of the th typical fault

[0081] In step S6, the calculated comprehensive fault current signal satisfies:

[0082]

[0083] where, represents the value of the original current signal at the sampling index when an unknown fault occurs, represents the value of the comprehensive current signal at the discrete index when an unknown fault occurs.

[0084] In step S6, using the comprehensive fault current signal as a sample, the calculated fault prediction current obtained by the pre-trained model satisfies:

[0085]

[0086] where, to respectively represent the fault prediction currents collected at the first to the (N - 1)-th sampling moments obtained by the pre-trained model when an unknown fault occurs in the motor; represents the pre-trained weight coefficient matrix, which is a row column matrix, represents the pre-trained weight coefficient at the th row and the th column, and so on.

[0087] In step S6, the constructed fault prediction current vector satisfies:

[0088]

[0089] where, represents the fault prediction current vector when an unknown fault occurs in the motor.

[0090] Calculate the vector difference between the fault prediction current vector and the average current vector of any one of the typical faults to satisfy:

[0091]

[0092] where, represents the vector difference between the fault prediction current vector when an unknown fault occurs in the motor and the average current vector of the th typical fault.

[0093] In step S7, select the typical fault with the smallest vector difference as the determined fault to satisfy:

[0094]

[0095] where, Fault qualitative diagnosis result of the unknown motor fault indicated; Indicates the vector difference between the fault prediction current vector and the typical fault average current vector Obtains the minimum value of the fault type.

[0096] Specifically, motor faults can be divided into various types, including bearing faults, insulation aging, winding faults, brush wear, overload and overheating, fan faults, etc. These faults not only affect the normal operation of the motor but may also cause more serious damage;

[0097] ‌Fault location and specific reasons‌ include:

[0098] I. ‌Excessive bearing operating temperature‌: It may be due to incorrect lubricant grade, deterioration, excessive or insufficient amount, bearing wear, rust, pitting, inner or outer race running, poor assembly of inner and outer small covers or oil retaining rings, misalignment of the coupling or too tight belt tension, etc.

[0099] II. ‌Excessive temperature or smoking‌: It may be due to problems with the power supply, such as too high, too low or missing phase voltage; problems with the motor itself, such as interphase, turn - to - turn short - circuit or grounding of the stator winding, broken rotor bars or rubbing between the stator and rotor; problems with the load, such as excessive or jammed mechanical load; problems with ventilation and heat dissipation, such as high ambient temperature, dirty outer shell, blocked air duct, damaged or wrongly installed fan, etc.

[0100] III. ‌Abnormal noise or strong vibration‌: It may be due to rubbing between the stator and rotor or severe wear and deformation of the driven machinery, uneven foundation, unstable foundation or loose anchor bolts, misalignment of the coupling or bent shaft, rotor eccentricity or imbalance, lack of oil or damage to the bearing, etc.

[0101] Embodiment 2

[0102] The present invention also provides a method for qualitative diagnosis of motor system faults based on classification and analysis of current sample characteristics, and its process is as follows:

[0103] 1) Current signal acquisition, using sensors to collect current and voltage data during motor operation, and these data will be used for subsequent feature extraction and diagnosis. ‌The specific current signal acquisition is as follows:

[0104] First, connect the motor driver and correctly connect the current sensor to the power line;

[0105] Secondly, collect the current signal, and use the analog circuit and sampling chip in the driver to collect and convert the current signal of the connected current sensor;

[0106] Finally, after converting the collected current signal into digital form, data processing and analysis can be carried out.

[0107] It should be noted that the sensor also collects voltage data and temperature data of the motor, and applies them to the diagnosis of current faults based on the characteristics of multi-source data fusion. According to the extraction of multi-source data fusion features, faults are identified and predicted through artificial intelligence models and methods.

[0108] 2) Data preprocessing: Preprocess the collected current data, including operations such as noise removal and normalization to ensure the accuracy and reliability of the data, and obtain a sample data set.

[0109] Specifically, it includes:

[0110] Data cleaning, including handling missing values and outliers. Missing values can be processed by methods such as deletion, replacement, or imputation; outliers are identified and processed through statistical methods and machine learning methods.

[0111] Data integration: Merge data from different sources, and it is necessary to solve problems of data conflicts and inconsistencies.

[0112] Data reduction: Compress data through algorithms or reduce the dimensionality of data using clustering methods to improve processing efficiency and accuracy.

[0113] Furthermore, the preprocessing also includes obtaining the stator current signal, reconstructing the noise signal, and then subtracting the reconstructed signal from the original stator current signal to obtain the residual current. The noise signal reconstruction formula is:

[0114] r(t)=s(t)+n(t)

[0115] Where r(t) is the reconstructed signal, s(t) is the original signal, and n(t) is the noise signal.

[0116] 3) Establish a deep learning model

[0117] Based on the convolutional neural network (CNN), CNN can identify abnormal patterns in current data and reduce false alarms and missed alarms by optimizing the model. Through analyzing the morphological features of current data, intelligent diagnosis of motor faults is realized.

[0118] The neural network algorithm diagnoses the normal and various abnormal states of the motor, including problems such as bearing damage, inter-turn short circuit, rotor bar breakage, and fault diagnosis.

[0119] The deep learning model requires a large number of labeled samples for training, so it is necessary to label the fault samples.

[0120] As a non-invasive diagnostic method, the motor current purely reflects the operating conditions of the motor. Fault diagnosis based on motor current signature analysis (MCSA) has the following advantages: a) The current signal is easy to collect, only by passing the power line through the current sensor; b) Signal acquisition does not interfere with the system operation; c) The collected signal is robust to environmental noise; d) The current sensor is inexpensive.

[0121] 4) Fault feature extraction

[0122] Specifically, first build a simulation in SIMULINK to obtain fault data, and then perform fault diagnosis based on signal processing feature extraction classification and recognition algorithms;

[0123] The working principle of the motor is essentially the interaction between electricity and magnetism. When the motor is abnormal, the original electromagnetic balance will be broken, and the fault features will be reflected in the stator circuit in the form of specific harmonics. Motor fault monitoring based on current signature analysis is to identify the characteristic signals related to motor faults in the spectrum by analyzing the motor current signal.

[0124] This embodiment is carried out by means of preprocessing. First of all, data preprocessing is an important step in fault feature extraction. It is necessary to clean the original data, remove outliers and noise, etc. to ensure the accuracy and reliability of the data. This step can be carried out using tools such as MATLAB and Python.

[0125] 5) Fault type identification and diagnosis

[0126] Through the trained deep learning model, the motor signal can be fault-identified and located. According to the characteristics of the input signal, the model can judge whether there is a fault in the motor and locate the specific position of the fault; and the deep learning-based method can also realize fault diagnosis and prediction. By analyzing the signal characteristics generated when the motor fails, the cause of the fault can be diagnosed and the possible future faults can be predicted.

[0127] In addition, by detecting the amplitude and waveform of the load current and spectrum analysis, the cause and degree of the motor fault are diagnosed. For example, by detecting the current of the motor and performing spectrum analysis to diagnose whether there are defects such as broken rotor bars, air gap eccentricity, stator winding faults, and rotor imbalance in the motor.

[0128] Furthermore, it also includes using a classifier to qualitatively diagnose motor faults, including: through the training of the simulation model and actual data, the CNN can learn the feature representation of the fault, so as to realize fault diagnosis.

[0129] The core advantages of the motor system fault qualitative diagnosis method based on the classification analysis of current sample features of the present invention lie in its high efficiency and accuracy. This method can monitor the health status of the motor in real time by analyzing the current signal during motor operation, and give early warnings before faults occur, thereby improving the reliability and safety of motor operation.

[0130] It can be understood that the present invention is described through some embodiments. Those skilled in the art know that without departing from the spirit and scope of the present invention, various changes or equivalent replacements can be made to these features and embodiments. Additionally, under the teaching of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.

Claims

1. A motor system fault qualitative diagnosis method based on current sample feature classification analysis, characterized in that: The following steps are involved: S1. Obtaining a comprehensive current signal and a fault current signal: performing multi-point sampling of current samples when a fault occurs during the operation of the motor system through multiple sensors to obtain an original current signal, and converting the original current signal into a comprehensive current signal; performing time-sharing sampling of the motor current when the motor fails to obtain a fault current signal; The current signal acquisition is specifically as follows: S11, connect the motor driver and connect the current sensor to the power line; S12, collecting current signals, using the sampling circuit and sampling chip in the driver to sample the current signals of the connected current sensors to obtain the original current signals; S13, converting the collected original current signal into a comprehensive current signal, and the conversion formula satisfies: ； in, Represents the discrete index of the integrated current signal, which is used to refer to the output sampling point calculated by the conversion formula. Represents a discrete index The value of the integrated current signal; represents the sampling index of the original signal, Indicates the sampling index The value of the original current signal; is the imaginary number symbol, is the natural base; S14, sampling the motor current in time when the motor fails to obtain a fault current signal ,satisfy: ； in, Respectively represent the first to N-1th sampling moments; S2. Construct the comprehensive current signal and the fault current signal into a pre-training data set; construct the pre-training samples to meet the following requirements: ； in, to They respectively represent the fault current signals collected from the first to the N-1th sampling moments when a motor fault occurs; represents the weight coefficient matrix, which is OK Column matrix, Indicates Line The weight coefficient of the column, and so on; S3. Complete the construction of the pre-training model based on the pre-training data set: establish a deep learning model based on the convolutional neural network, import the processed pre-training data set into the model for training, obtain the pre-trained weight coefficient matrix, and complete the construction of the pre-training model; S4. Complete fault qualitative classification: perform fault qualitative classification on the fault samples based on fault type, fault degree and fault location information; S5. Obtaining a typical fault average current vector: establishing a fault current vector for the fault current signal of the fault sample after the fault qualitative classification, and obtaining a typical fault average current vector; The fault current vector satisfies: ； in, Indicates Typical fault The fault current vector of samples, Indicates Typical fault Samples in The sampling current at the moment, and so on; The typical fault average current vector satisfies: ； in, Indicates Typical fault average current vector of typical faults, Indicates The total number of samples of typical faults; S6. Constructing a fault prediction current vector: sampling the current of the motor system in real time, obtaining a fault sampling current when an unknown fault occurs, calculating a fault comprehensive current signal, using the fault comprehensive current signal as a sample, using a pre-trained model to calculate a fault prediction current, and constructing a fault prediction current vector; The calculated fault comprehensive current signal satisfies: ； in, Indicates the sampling index when an unknown fault occurs The value of the original current signal, Indicates a discrete index when an unknown fault occurs The value of the integrated current signal; The fault comprehensive current signal is used as a sample and the pre-trained model is used to calculate the fault prediction current to satisfy: ； in, to They respectively represent the fault prediction currents collected from the first to the N-1th sampling moments obtained by the pre-trained model when an unknown fault occurs in the motor; Represents the weight coefficient matrix after pre-training, which is OK Column matrix, Indicates Line The pre-trained weight coefficients of the column, and so on; Constructing the fault prediction current vector satisfies: ； in, Represents the fault prediction current vector when an unknown fault occurs in the motor; S7. Complete qualitative fault diagnosis for the motor system: calculate the vector difference between the fault prediction current vector and any typical fault average current vector, select a typical fault with the smallest vector difference as the judgment fault, and complete qualitative fault diagnosis for the motor system.

2. The motor system fault qualitative diagnosis method based on current sample feature classification analysis according to claim 1 is characterized in that: In step S7, the vector difference between the fault prediction current vector and any typical fault average current vector is calculated to satisfy: ； in, The fault prediction current vector when the motor has an unknown fault is The vector difference of the average current vector of typical faults.

3. The motor system fault qualitative diagnosis method based on current sample feature classification analysis according to claim 2 is characterized in that: In step S7, a typical fault with the smallest vector difference is selected as a judgment fault that satisfies: ； in, The qualitative fault diagnosis result of the motor unknown fault is shown; Represents the vector difference between the fault prediction current vector and the typical fault average current vector Get the minimum value The fault type.

Images