WO2020078132A1 - 一种电机轴承故障诊断装置 - Google Patents

一种电机轴承故障诊断装置 Download PDF

Info

Publication number
WO2020078132A1
WO2020078132A1 PCT/CN2019/104202 CN2019104202W WO2020078132A1 WO 2020078132 A1 WO2020078132 A1 WO 2020078132A1 CN 2019104202 W CN2019104202 W CN 2019104202W WO 2020078132 A1 WO2020078132 A1 WO 2020078132A1
Authority
WO
WIPO (PCT)
Prior art keywords
bearing
fault
signal
residual signal
frequency
Prior art date
Application number
PCT/CN2019/104202
Other languages
English (en)
French (fr)
Inventor
朱文龙
刘勇
戴计生
李宗帅
杨家伟
江平
詹彦豪
张中景
徐海龙
Original Assignee
株洲中车时代电气股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 株洲中车时代电气股份有限公司 filed Critical 株洲中车时代电气股份有限公司
Priority to US17/261,886 priority Critical patent/US11898932B2/en
Priority to EP19872841.2A priority patent/EP3816604B1/en
Publication of WO2020078132A1 publication Critical patent/WO2020078132A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/34—Testing dynamo-electric machines
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/0092—Arrangements for measuring currents or voltages or for indicating presence or sign thereof measuring current only
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
    • G01R23/16—Spectrum analysis; Fourier analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the invention relates to the technical field of fault detection, in particular to a device for fault diagnosis of a traction motor bearing using a current signal.
  • the traction motor In electric railway locomotives, the traction motor is the most important component to realize the conversion of electrical energy and mechanical energy. It bears the important task of power output during the operation of locomotives and has an important impact on the operating efficiency and safety of railway locomotives. Operation practice shows that the motor bearing failure is the most common and dangerous failure of the traction motor. The occurrence and development of these failures not only cause damage to the motor, but may also cause damage to other equipment, resulting in great losses. How to carry out timely and effective state monitoring and fault diagnosis of traction motor bearing failures to avoid economic losses caused by vicious accidents and unnecessary shutdowns is the key technical problem to solve the state maintenance of traction motors.
  • the vibration signal analysis method and the stator current signal analysis method are the two most widely used methods in the field of traction motor bearing fault diagnosis.
  • additional equipment such as vibration sensors need to be added, which not only increases the cost but also brings new safety risks.
  • the stator current signal analysis method is a non-intrusive fault diagnosis method, that is, it does not require additional sensors, and has the advantages of low cost, low cost, and easy implementation.
  • PWM pulse width modulation
  • MCSA motor current characteristic analysis
  • Prior art 1 was applied by Anhui University on December 23, 2016, and published on May 31, 2017.
  • the Chinese invention application number CN106769041A "Online fault of a permanent magnet synchronous motor bearing under variable speed conditions" Diagnostic device and method ".
  • the invention application discloses a device and method for online fault diagnosis of permanent magnet synchronous motor bearings under variable speed working conditions.
  • the analog-to-digital converter 1 is connected to a current probe and collects current signals at equal intervals, and the microcontroller 1 performs low-pass filtering and polarity conversion on the current signals.
  • the angle of the converted unipolar current signal is calculated and rounded. When the angle changes by 1 degree, the microcontroller 1 generates a trigger signal.
  • the micro-controller 2 receives the trigger signal generated by the micro-controller 1 and controls the analog-to-digital converter 2 to trigger sampling of the microphone to obtain the bearing sound signal. Envelope demodulation of the bearing signal in the angular domain, calculate the order spectrum of the envelope signal, determine the bearing fault type according to the fault feature order, and display it on the display screen. According to the analysis and processing of the sound signal of the bearing, the invention judges the bearing failure type according to the failure feature order, and needs to add additional sensors, analog-to-digital converters and other components, which is not conducive to cost reduction. At the same time, it is difficult to effectively extract fault characteristic signals from current signals with low signal-to-noise ratio by using bearing sound signals to determine bearing fault types.
  • the prior art 2 was applied by Nanjing University of Aeronautics and Astronautics on March 28, 2017, and published on July 28, 2017.
  • the Chinese invention application for the publication number CN106989923A "Permanent Magnetic Machine Bearing Pitting Based on Wavelet Packet Analysis of Stator Current Fault detection method.
  • the invention application discloses a method for detecting the pitting corrosion of the bearing of a permanent magnet machine based on the analysis of the stator current wavelet packet.
  • the stator current of the permanent magnet machine is used for fault diagnosis, and the stator current signal is analyzed by the wavelet packet analysis method.
  • the wavelet packet analysis is a multi-resolution analysis method, which can strip the signal according to the frequency band.
  • the corresponding wavelet packet node is calculated, the root mean square of the wavelet packet node coefficient is calculated, and the fault condition is judged according to the root mean square.
  • the invention adopts wavelet packet analysis method to analyze the signal of the stator current, and judges the fault condition according to the root mean square of the wavelet packet node coefficient. It also has the technical problem that it is difficult to effectively extract the fault characteristic signal from the current signal with low signal-to-noise ratio.
  • the invention application discloses a method for fault diagnosis of permanent magnet motor bearings based on sensorless sensors. First, obtain the voltage and current from the motor control chip, and use the sensorless algorithm to obtain the rotor position angle and speed of the motor. Then, the AC component in the rotation speed is extracted through the moving average filter, and it is resampled in the angular domain according to the position angle. Finally, the frequency domain analysis is performed on the resampled speed signal, and the bearing fault is judged based on the frequency domain information.
  • the invention needs to obtain the rotor position angle and rotation speed of the motor according to the voltage and current through the sensorless algorithm, resample the angular domain according to the position angle, and then resample the angular domain according to the position angle. Not only is the algorithm complex, but it also has the technical problem that it is difficult to effectively extract the fault characteristic signal from the low signal-to-noise ratio current signal.
  • the object of the present invention is to provide a motor bearing fault diagnosis device to solve the difficulty of the existing motor bearing fault diagnosis device to effectively extract the fault characteristic signal from the low signal-to-noise ratio current signal, resulting in false alarms or Underreported technical issues.
  • a motor bearing fault diagnosis device includes: a current signal acquisition unit, a residual signal acquisition unit, a fault feature extraction unit and a fault Diagnostic model unit.
  • a residual signal acquisition unit extracting harmonic signals from the real-time motor current signal acquired by the current signal acquisition unit, eliminating the fundamental frequency and harmonic signals from the original real-time motor current signal, and obtaining a residual signal;
  • the fault feature extraction unit performs time and frequency domain analysis on the residual signal obtained by the residual signal acquisition unit to extract the fault feature index of the bearing;
  • the fault diagnosis model unit uses the trained bearing fault diagnosis model to perform pattern recognition on the fault feature index of the bearing extracted by the fault feature extraction unit, and diagnoses the fault state of the bearing.
  • the fault feature extraction unit first calculates the bearing fault feature frequency, and then calculates the bearing fault feature frequency sequence. Carry out FFT transformation on the residual signal to obtain the residual signal spectrum, and then obtain the corresponding bearing fault characteristic frequency sequence amplitude according to the residual signal spectrum. Calculate the minimum value I_min and effective value I_rms of the residual signal, and calculate the maximum value I_f_max, peak-to-peak value I_f_peark and peak coefficient I_f_crest of the bearing fault characteristic frequency sequence amplitude I_f (n).
  • the current signal acquisition unit acquires multiple sets of motor current history signals under different bearing fault conditions as the training sample data set of the model
  • a residual signal acquiring unit extracting harmonic signals from the training sample data set acquired by the current signal acquiring unit, eliminating the fundamental frequency and harmonic signals from the original motor current history signal, and obtaining a residual signal;
  • the fault feature extraction unit performs time and frequency domain analysis on the residual signal obtained by the residual signal acquisition unit to extract the fault feature index of the bearing;
  • the fault diagnosis model unit is based on the fault feature index of the bearing extracted by the fault feature extraction unit and combined with the bearing fault type for training to obtain a bearing fault diagnosis model.
  • the residual signal acquisition unit obtains the amplitude, initial phase and frequency of the fundamental frequency and harmonics of the motor current history signal from the original motor current history signal, and Obtain the fundamental frequency and harmonic signals, and then subtract the fundamental frequency and harmonic signals from the original motor current history signal to obtain the residual signal.
  • the residual signal acquisition unit obtains the amplitude, initial phase and frequency of the fundamental frequency and harmonics of the real-time signal of the motor current from the real-time signal of the motor current to be diagnosed, and Obtain the fundamental frequency and harmonic signals, and then subtract the fundamental frequency and harmonic signals from the original real-time signal of the motor current to obtain the residual signal.
  • the fault feature extraction unit first calculates the bearing fault feature frequency, and then calculates the bearing fault feature frequency sequence. Carry out FFT transformation on the residual signal to obtain the residual signal spectrum, and then obtain the corresponding bearing fault characteristic frequency sequence amplitude according to the residual signal spectrum. Calculate the minimum value I_min and effective value I_rms of the residual signal, and calculate the maximum value I_f_max, peak-to-peak value I_f_peark and peak coefficient I_f_crest of the bearing fault characteristic frequency sequence amplitude I_f (n).
  • the residual signal acquisition unit uses the full-phase time-shifted phase difference correction method to obtain the amplitude, initial phase and frequency of the fundamental frequency and harmonics of the motor current history signal, In turn, the fundamental frequency and harmonic signals are obtained.
  • the residual signal acquisition unit uses the full-phase time-shifted phase difference correction method to obtain the amplitude, initial phase and frequency of the fundamental frequency and harmonics of the real-time signal of the motor current, thereby obtaining the fundamental frequency And harmonic signals.
  • the fault diagnosis model unit uses a gradient lifting tree classifier to train the extracted fault feature indexes and bearing fault types of the bearing to obtain a bearing fault diagnosis model.
  • the fault diagnosis model unit uses a trained bearing fault diagnosis model based on a gradient lifting tree to perform pattern recognition on the extracted fault feature indexes of the bearing to diagnose the fault state of the bearing.
  • the characteristic frequency sequence f (k) of the bearing failure is
  • , k 1, 2, 3, where fe is the current fundamental frequency of the residual signal, and fv is the characteristic frequency of the bearing failure.
  • the fault feature extraction unit calculates the minimum value I_min of the residual signal according to the following formula:
  • the effective value I_rms of the residual signal is calculated according to the following formula:
  • I_f_max max (I_f (n))
  • I_f (n) is the amplitude of the characteristic frequency sequence of the bearing fault.
  • I_f_peark I_f_max-I_min
  • I_f_crest I_f_max / I_rms.
  • bearing failure states include bearing cage failure, bearing outer ring failure, bearing inner ring failure and bearing rolling element failure.
  • the fault feature extraction unit calculates the fault feature frequency of the bearing cage according to the following formula:
  • the fault feature extraction unit calculates the fault feature frequency of the bearing outer ring according to the following formula:
  • the fault feature extraction unit calculates the fault feature frequency of the bearing inner ring according to the following formula:
  • the fault feature extraction unit calculates the fault feature frequency of the bearing rolling element according to the following formula:
  • fr is the rotational frequency of the motor
  • D B is the diameter of the rolling element bearing
  • D P is the pitch diameter of the bearing section
  • N B is the number of bearing rolling elements
  • ⁇ is the contact angle
  • the present invention adopts harmonic signal extraction of the motor current signal to eliminate the fundamental frequency and harmonic signals from the original motor current history signal to obtain a residual signal, which effectively reduces the PWM (Pulse Width Modulation) (Abbreviation of)
  • PWM Pulse Width Modulation
  • the harmonic signal introduced by the power supply interferes with the bearing fault characteristic signal, which improves the signal-to-noise ratio of the current signal, and the extracted fault characteristic index is more significant;
  • the present invention uses the change of the bearing fault feature index extracted by the residual signal to reflect the change of the health state of the traction motor bearing.
  • the fault feature index can accurately characterize the bearing fault state.
  • the full phase time shift phase difference correction method is used to extract and The harmonic signal in the current signal is removed, which further improves the significance of the fault characteristic index;
  • the gradient lifting tree fault diagnosis model established by the present invention can automatically identify the bearing fault status without the need for business experts to observe the spectrum chart for diagnosis, and can automatically identify the bearing fault according to the characteristics of the motor current signal, effectively reducing It reduces the interference of human subjective factors and improves the objectivity and accuracy of fault diagnosis.
  • FIG. 1 is a block diagram of the working principle of a specific embodiment of a motor bearing fault diagnosis device of the present invention in a fault diagnosis mode;
  • FIG. 2 is a block diagram of the working principle of a specific embodiment of a motor bearing fault diagnosis device of the present invention in a model training mode;
  • FIG. 3 is a schematic diagram of the flow principle of a method for diagnosing a motor bearing fault based on the device of the present invention
  • FIG. 4 is a program flowchart of a model training stage in a motor bearing fault diagnosis method based on the device of the present invention
  • Fig. 5 is a flowchart of a fault diagnosis phase in a fault diagnosis method of a motor bearing based on the device of the present invention
  • All-phase time-shift phase difference correction method It is a kind of interference elimination algorithm for sinusoidal signals in the signal, which is used to calculate the amplitude, frequency and initial phase of the fundamental frequency and harmonic signals.
  • Gradient lifting tree It is a machine learning algorithm and a member of the integrated learning Boosting family. It uses the error rate of the previous round of iterative weak learners to update the weight of the training set.
  • an embodiment of a bearing fault diagnosis device based on a current signal specifically includes: a current signal acquisition unit 1, a residual signal acquisition unit 2, a fault feature extraction unit 3, and a fault diagnosis model unit 4.
  • a current signal acquisition unit 1 a current signal acquisition unit 1
  • a residual signal acquisition unit 2 a residual signal acquisition unit 2
  • a fault feature extraction unit 3 a fault diagnosis model unit 4.
  • the current signal acquisition unit 1 acquires the real-time signal of the motor current
  • the residual signal acquisition unit 2 performs harmonic signal extraction on the real-time motor current signal acquired by the current signal acquisition unit 1, and removes the fundamental frequency and harmonic signals from the original real-time motor current signal to obtain a residual signal;
  • the fault feature extraction unit 3 performs time domain and frequency domain analysis on the residual signal obtained by the residual signal acquisition unit 2 to extract the fault feature index of the bearing;
  • the fault diagnosis model unit 4 uses the trained bearing fault diagnosis model to perform pattern recognition on the fault feature index (vector) of the bearing extracted by the fault feature extraction unit 3, and diagnoses the fault state of the bearing (ie, the type of bearing fault).
  • the fault diagnosis model unit 4 inputs the extracted fault feature indexes (vectors) of the bearings into the trained gradient-based lifting tree (the algorithm is already in the prior art). There is an algorithm, which will not be repeated here.)
  • the bearing fault diagnosis model carries out pattern recognition to diagnose the fault state of the bearing.
  • the fault feature extraction unit 3 first calculates the bearing fault feature frequency, and then calculates the bearing fault feature frequency sequence.
  • the residual signal is subjected to FFT (Fast Fourier Transformation, short for Fast Fourier Transform) to obtain the frequency spectrum of the residual signal, and then the corresponding amplitude value of the characteristic frequency sequence of the bearing fault is obtained according to the frequency spectrum of the residual signal.
  • FFT Fast Fourier Transformation, short for Fast Fourier Transform
  • the fault diagnosis model unit 4 uses the trained bearing fault diagnosis model to extract the fault feature frequency sequence amplitude I_f (n), the maximum value of the bearing fault feature frequency sequence amplitude I_f_max, the peak-to-peak value extracted by the fault feature extraction unit 3 I_f_peark and the peak coefficient I_f_crest, as well as the residual error signal minimum value I_min and effective value I_rms of the bearing's fault characteristic index (vector) for pattern recognition, diagnose the fault state of the bearing.
  • the current signal acquisition unit 1 acquires multiple sets of motor current history signals under different bearing failure states (corresponding to different bearing failure types) as a training sample data set of the model;
  • the residual signal acquisition unit 2 performs harmonic signal extraction on the training sample data set acquired by the current signal acquisition unit 1, and removes the fundamental frequency and harmonic signals from the original motor current history signal to obtain a residual signal;
  • the fault feature extraction unit 3 performs time domain and frequency domain analysis on the residual signal obtained by the residual signal acquisition unit 2 to extract the fault feature index of the bearing;
  • the fault diagnosis model unit 4 is based on the fault feature index (vector) of the bearing extracted by the fault feature extraction unit 3 and combined with the bearing fault type for training to obtain a bearing fault diagnosis model.
  • the fault diagnosis model unit 4 inputs the extracted fault feature index (vector) of the bearing and the bearing fault type to the gradient lifting tree classifier (the algorithm is in the prior art The existing algorithm of, will not be repeated here), and the bearing fault diagnosis model is obtained.
  • this algorithm is an existing algorithm in the prior art, which will not be repeated here
  • the fault feature extraction unit 3 When the device is in a fault model training state, the fault feature extraction unit 3 first calculates the bearing fault feature frequency, and then calculates the bearing fault feature frequency sequence.
  • Bearing fault characteristic frequency sequence f (k) is further
  • , k 1, 2, 3, ..., n, where n can be taken 3 to 5 times, fe is the current fundamental frequency of the residual signal , Fv is the characteristic frequency of bearing failure.
  • Fault diagnosis model unit 4 based on the fault feature extraction unit 3 extracted under different bearing fault conditions, including the bearing fault characteristic frequency sequence amplitude I_f (n), the maximum value of the bearing fault characteristic frequency sequence amplitude I_f_max, the peak-to-peak value I_f_peark and The peak coefficient I_f_crest, and the residual characteristic signal minimum value I_min and effective value I_rms of the bearing fault characteristic index (vector), combined with the bearing fault type for training, get the bearing fault diagnosis model.
  • the fault feature extraction unit 3 further calculates the minimum value I_min of the residual signal according to the following formula:
  • the fault feature extraction unit 3 further calculates the effective value I_rms of the residual signal according to the following formula:
  • the fault feature extraction unit 3 further calculates the maximum value of the amplitude of the bearing fault feature frequency sequence according to the following formula:
  • I_f_max max (I_f (n))
  • I_f (n) is the amplitude of the characteristic frequency sequence of the bearing fault.
  • the fault feature extraction unit 3 further calculates the peak-to-peak value of the magnitude of the bearing fault feature frequency sequence according to the following formula:
  • I_f_peark I_f_max-I_min
  • the fault feature extraction unit 3 further calculates the peak coefficient of the magnitude of the bearing fault feature frequency sequence according to the following formula:
  • I_f_crest I_f_max / I_rms
  • bearing failure states further include bearing cage failure, bearing outer ring failure, bearing inner ring failure and bearing rolling element failure.
  • the fault feature extraction unit 3 further calculates the fault feature frequency of the bearing cage according to the following formula:
  • the fault feature extraction unit 3 further calculates the fault feature frequency of the bearing outer ring according to the following formula:
  • the fault feature extraction unit 3 further calculates the fault feature frequency of the bearing inner ring according to the following formula:
  • the fault feature extraction unit 3 further calculates the fault feature frequency of the bearing rolling body according to the following formula:
  • fr is the rotational frequency of the motor
  • D B is the diameter of the rolling element bearing
  • D P is the pitch diameter of the bearing section
  • N B is the number of bearing rolling elements
  • ⁇ is the contact angle
  • the fault diagnosis model unit 4 uses a gradient lifting tree intelligent algorithm for bearing fault diagnosis model training and pattern recognition. Neural networks, random forests, and support vector machines can also be used Alternative processing such as artificial intelligence algorithms.
  • the motor bearing fault diagnosis device described in Embodiment 1 can be well applied to offline or online health monitoring and fault early warning of locomotive traction motor bearing faults.
  • an embodiment of a bearing fault diagnosis method based on the device described in Embodiment 1 specifically includes the following steps:
  • step S103 Perform time and frequency domain analysis on the residual signal obtained in step S102) to extract the fault characteristic index of the bearing;
  • step S202 Perform time domain and frequency domain analysis on the residual signal obtained in step S201) to extract the fault characteristic index of the bearing;
  • step S203 input the fault feature index (vector) of the bearing extracted in step S202) to the bearing fault diagnosis model trained in step S104) to perform pattern recognition, and diagnose the fault state of the bearing.
  • the motor bearing fault diagnosis device described in the specific embodiment of the present invention adopts harmonic signal extraction of the motor current signal to eliminate the fundamental frequency and harmonic signals from the original motor current history signal to obtain a residual signal and effectively reduce
  • the harmonic signal introduced by the PWM power supply interferes with the bearing fault characteristic signal, improves the signal-to-noise ratio of the current signal, and the extracted fault characteristic index is more significant;
  • the motor bearing fault diagnosis device described in the specific embodiment of the present invention uses the change of the bearing fault feature index extracted by the residual signal to reflect the change of the traction motor bearing health state, and the fault feature index can accurately characterize the bearing fault state,
  • the full-phase time shift phase difference correction method is used to extract and remove the harmonic signal in the current signal, which further improves the significance of the fault characteristic index;
  • the established gradient lifting tree fault diagnosis model can automatically identify the bearing fault state without the need for business experts to observe the spectrum chart for diagnosis, and can be based on the current signal Features automatically identify whether the bearing is faulty, effectively reducing the interference of human subjective factors, and improving the objectivity and accuracy of fault diagnosis.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Acoustics & Sound (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

一种电机轴承故障诊断装置，当装置处于故障诊断状态时，电流信号获取单元获取电机电流实时信号。残差信号获取单元对电流信号获取单元获取的电机电流实时信号进行谐波信号提取，从原始的电机电流实时信号中消除基频和谐波信号，获得残差信号。故障特征提取单元对残差信号获取单元得到的残差信号进行时域和频域分析，提取轴承的故障特征指标。故障诊断模型单元利用训练好的轴承故障诊断模型对故障特征提取单元提取到的轴承的故障特征指标进行模式识别，诊断出该轴承的故障状态。本发明能够解决现有电机轴承故障诊断装置难以有效地从低信噪比的电流信号中提取出故障特征信号，从而导致误报或漏报的技术问题。

Description

一种电机轴承故障诊断装置
本申请要求于2018年10月15日提交中国专利局、申请号为201811197298.4、发明名称为“一种电机轴承故障诊断装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及故障检测技术领域，尤其是涉及一种采用电流信号进行牵引电机轴承故障诊断的装置。
背景技术
在电力铁路机车车辆中，牵引电机是实现电能和机械能转换的最核心部件，承担了机车车辆运行中动力输出的重任，对于铁路机车车辆的运行效率与安全性具有重要的影响。运行实践表明，电机轴承故障是牵引电机最常见和最危险的故障，这些故障的发生和发展，不仅导致电机损坏，而且可能引起其它设备损坏，从而造成很大的损失。如何对牵引电机轴承故障进行及时、有效的状态监测及故障诊断，以避免恶性事故和不必要的停机造成的经济损失是解决牵引电机状态维修的关键技术问题。
目前，基于振动信号分析法和定子电流信号分析法是牵引电机轴承故障诊断领域应用最为广泛的两种方法。采用振动信号分析法时，需要额外加装振动传感器等设备，这样不仅增加了成本，而且带来了新的安全隐患。而定子电流信号分析方法则是一种非侵入式故障诊断方法，即不需要额外增加传感器，具有低成低廉、易于实施等优点。然而，受到脉冲宽度调制(PWM)供电电源和工况复杂多变的影响，牵引电机电流信号中蕴含着丰富的电机运行状态信息，而轴承故障特征信号相对微弱，往往淹没在电流谐波等干扰信号中。传统的电机电流特征分析(MCSA)算法，如：傅里叶变换(FFT)、PARK矢量模分析等算法，均难以有效地从低信噪比的电流信号中提取出故障特征信号，从而导致误报或漏报情况的发生。
在现有技术中，主要有以下技术方案与本发明申请相关：
现有技术1为安徽大学于2016年12月23日申请，并于2017年05 月31日公开，公开号为CN106769041A的中国发明申请《一种变转速工况下的永磁同步电机轴承在线故障诊断装置及方法》。该发明申请公开了一种变转速工况下的永磁同步电机轴承在线故障诊断装置及方法。模数转换器1连接电流探头，等时间间隔采集电流信号，微控制器1对电流信号进行低通滤波和极性转换。计算转换后的单极性电流信号角度并取整，角度每变化1度时，微控制器1产生一个触发信号。微控制器2接收微控制器1产生的触发信号，控制模数转换器2对麦克风进行触发采样，获得轴承声音信号。对角域的轴承信号进行包络解调，计算包络信号的阶次谱，根据故障特征阶次判断轴承故障类型，并在显示屏上显示。该发明通过对轴承声音信号进行分析处理，根据故障特征阶次判断轴承故障类型，需要额外增加传感器、模数转换器等部件，不利于成本的降低。同时，采用轴承声音信号判断轴承故障类型难以有效地从低信噪比的电流信号中提取出故障特征信号。
现有技术2为南京航空航天大学于2017年03月28日申请，并于2017年07月28日公开，公开号为CN106989923A的中国发明申请《基于定子电流小波包分析的永磁机轴承点蚀故障检测方法》。该发明申请公开了一种基于定子电流小波包分析的永磁机轴承点蚀故障检测方法。利用永磁机的定子电流进行故障诊断，通过小波包分析方法对定子电流进行信号分析，小波包分析是一种多分辨率分析法，可以按频段对信号进行剥离。当确定电机故障频率后，计算出相应的小波包节点，对小波包节点系数求均方根，根据均方根判断故障情况。该发明采用小波包分析方法对定子电流进行信号分析，根据小波包节点系数的均方根判断故障情况，也存在难以有效地从低信噪比的电流信号中提取出故障特征信号的技术问题。
现有技术3微浙江大学于2017年01月20日申请，并于2017年05月17日公开，公开号为CN106680716A的中国发明申请《一种基于无位置传感器的永磁电机轴承故障诊断的方法》。该发明申请公开了一种基于无位置传感器的永磁电机轴承故障诊断的方法。首先，从电机控制芯片中获得电压与电流，并利用无位置传感器算法获得电机的转子位置角与转速。然后，通过滑动平均滤波器提取出转速中的交流成分，并根据位置角度对其 进行角域重采样。最后，对重采样速度信号进行频域分析，根据频域信息判断轴承故障是否发生。该发明需要根据电压与电流通过无位置传感器算法获得电机的转子位置角与转速，根据位置角度对其进行角域重采样，再根据位置角度对其进行角域重采样。不但算法复杂，而且也存在难以有效地从低信噪比的电流信号中提取出故障特征信号的技术问题。
发明内容
有鉴于此，本发明的目的在于提供一种电机轴承故障诊断装置，以解决现有电机轴承故障诊断装置难以有效地从低信噪比的电流信号中提取出故障特征信号，从而导致误报或漏报的技术问题。
为了实现上述发明目的，本发明具体提供了一种电机轴承故障诊断装置的技术实现方案，一种电机轴承故障诊断装置，包括：电流信号获取单元、残差信号获取单元、故障特征提取单元和故障诊断模型单元。当所述装置处于故障诊断状态时：
电流信号获取单元，获取电机电流实时信号；
残差信号获取单元，对所述电流信号获取单元获取的电机电流实时信号进行谐波信号提取，从原始的电机电流实时信号中消除基频和谐波信号，获得残差信号；
故障特征提取单元，对所述残差信号获取单元得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
故障诊断模型单元，利用训练好的轴承故障诊断模型对所述故障特征提取单元提取到的轴承的故障特征指标进行模式识别，诊断出该轴承的故障状态。
进一步的，当所述装置处于故障诊断状态时，所述故障特征提取单元首先计算轴承故障特征频率，再计算轴承故障特征频率序列。对残差信号进行FFT变换获得残差信号的频谱，再根据残差信号的频谱获得对应的轴承故障特征频率序列幅值。计算残差信号的最小值I_min和有效值I_rms，并根据轴承故障特征频率序列幅值I_f(n)，计算轴承故障特征频率序列幅值的最大值I_f_max、峰峰值I_f_peark和峰值系数I_f_crest。
进一步的，当所述装置处于故障模型训练状态时：
电流信号获取单元，获取在不同轴承故障状态下的多组电机电流历史信号，作为模型的训练样本数据集；
残差信号获取单元，对所述电流信号获取单元获取的训练样本数据集进行谐波信号提取，从原始的电机电流历史信号中消除基频和谐波信号，获得残差信号；
故障特征提取单元，对所述残差信号获取单元得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
故障诊断模型单元，基于所述故障特征提取单元提取到的轴承的故障特征指标并结合轴承故障类型进行训练，得到轴承故障诊断模型。
进一步的，当所述装置处于故障模型训练状态时，所述残差信号获取单元从原始的电机电流历史信号中获得电机电流历史信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号，再从原始的电机电流历史信号中减去基频和谐波信号，从而获得残差信号。
进一步的，当所述装置处于故障诊断状态时，所述残差信号获取单元从待诊断的电机电流实时信号中获得电机电流实时信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号，再从原始的电机电流实时信号中减去基频和谐波信号，从而获得残差信号。
进一步的，当所述装置处于故障模型训练状态时，所述故障特征提取单元首先计算轴承故障特征频率，再计算轴承故障特征频率序列。对残差信号进行FFT变换获得残差信号的频谱，再根据残差信号的频谱获得对应的轴承故障特征频率序列幅值。计算残差信号的最小值I_min和有效值I_rms，并根据轴承故障特征频率序列幅值I_f(n)，计算轴承故障特征频率序列幅值的最大值I_f_max、峰峰值I_f_peark和峰值系数I_f_crest。
进一步的，当所述装置处于故障模型训练状态时，所述残差信号获取单元采用全相位时移相位差校正法获得电机电流历史信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号。当所述装置处于故障诊断状态时，所述残差信号获取单元采用全相位时移相位差校正法获得电机电流实时信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号。
进一步的，当所述装置处于故障模型训练状态时，所述故障诊断模型单元利用梯度提升树分类器对提取到的轴承的故障特征指标及轴承故障类型进行训练，得到轴承故障诊断模型。当所述装置处于故障诊断状态时，所述故障诊断模型单元利用训练好的基于梯度提升树的轴承故障诊断模型对提取到的轴承的故障特征指标进行模式识别，诊断出轴承的故障状态。
进一步的，所述轴承故障特征频率序列f(k)为|fe±kfv|,k＝1,2,3，其中，fe为残差信号的电流基频频率，fv为轴承故障特征频率。
进一步的，所述故障特征提取单元根据以下公式计算所述残差信号的最小值I_min：
I_min＝min(I(t))
式中，I(t)为残差信号的时间序列，其中t＝1～N，N为信号的采样点数。
根据以下公式计算所述残差信号的有效值I_rms：
Figure PCTCN2019104202-appb-000001
根据以下公式计算所述轴承故障特征频率序列幅值的最大值：
I_f_max＝max(I_f(n))
式中，I_f(n)为轴承故障特征频率序列幅值。
根据以下公式计算所述轴承故障特征频率序列幅值的峰峰值：
I_f_peark＝I_f_max-I_min
根据以下公式计算所述轴承故障特征频率序列幅值的峰值系数：
I_f_crest＝I_f_max/I_rms。
进一步的，所述轴承的故障状态类型包括轴承保持架故障、轴承外圈故障、轴承内圈故障和轴承滚动体故障。
所述故障特征提取单元根据以下公式计算所述轴承保持架故障特征频率：
Figure PCTCN2019104202-appb-000002
所述故障特征提取单元根据以下公式计算所述轴承外圈故障特征频率：
Figure PCTCN2019104202-appb-000003
所述故障特征提取单元根据以下公式计算所述轴承内圈故障特征频率：
Figure PCTCN2019104202-appb-000004
所述故障特征提取单元根据以下公式计算所述轴承滚动体故障特征频率：
Figure PCTCN2019104202-appb-000005
其中，fr为电机旋转频率，D B为轴承中滚动体的直径，D P为轴承节的圆直径，N B为轴承中滚动体的个数，θ为接触角。
通过实施上述本发明提供的电机轴承故障诊断装置的技术方案，具有如下有益效果：
(1)本发明采用对电机电流信号进行谐波信号提取，从原始的电机电流历史信号中消除基频和谐波信号，获得残差信号，有效减少了由PWM(Pulse Width Modulation，脉宽调制的简称)供电引入的谐波信号对轴承故障特征信号的干扰，提高了电流信号的信噪比，进而提取的故障特征 指标更加显著；
(2)本发明利用残差信号提取的轴承故障特征指标的变化来反映牵引电机轴承健康状态的变化，该故障特征指标能够准确地表征轴承故障状态，采用全相位时移相位差校正法提取并去除电流信号中的谐波信号，进一步提升了故障特征指标的显著性；
(3)本发明所建立的梯度提升树故障诊断模型可自动识别轴承故障状态，而不需要业务专家观察频谱图进行诊断，能够根据电机电流信号的自身特性自动地识别轴承故障与否，有效减少了人为主观因素的干扰，并提高了故障诊断的客观性和准确性。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单的介绍。显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的实施例。
图1是本发明电机轴承故障诊断装置一种具体实施例在故障诊断模式下的工作原理框图；
图2是本发明电机轴承故障诊断装置一种具体实施例在模型训练模式下的工作原理框图；
图3是基于本发明装置的电机轴承故障诊断方法的流程原理示意图；
图4是基于本发明装置的电机轴承故障诊断方法中模型训练阶段的程序流程图；
图5是基于本发明装置的电机轴承故障诊断方法中故障诊断阶段的程 序流程图；
图中：1-电流信号获取单元，2-残差信号计算单元，3-故障特征提取单元，4-故障诊断模型单元。
具体实施方式
为了引用和清楚起见，将下文中使用的技术名词、简写或缩写记载如下：
全相位时移相位差校正法：是一种消除信号中的正弦信号干扰算法，用于计算基频和谐波信号的幅值、频率和初始相位。
梯度提升树：是一种机器学习算法，也是集成学习Boosting家族的成员，是利用前一轮迭代弱学习器的误差率来更新训练集的权重。
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整的描述。显然，所描述的实施例仅仅是本发明的一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
如附图1至附图5所示，给出了本发明电机轴承故障诊断装置及基于该装置的电机轴承故障诊断方法的具体实施例，下面结合附图和具体实施例对本发明作进一步说明。
实施例1
如附图1所示，一种基于电流信号的轴承故障诊断装置的实施例，具体包括：电流信号获取单元1、残差信号获取单元2、故障特征提取单元3 和故障诊断模型单元4。当装置处于故障诊断状态时：
电流信号获取单元1，获取电机电流实时信号；
残差信号获取单元2，对电流信号获取单元1获取的电机电流实时信号进行谐波信号提取，从原始的电机电流实时信号中消除基频和谐波信号，获得残差信号；
故障特征提取单元3，对残差信号获取单元2得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
故障诊断模型单元4，利用训练好的轴承故障诊断模型对故障特征提取单元3提取到的轴承的故障特征指标(向量)进行模式识别，诊断出该轴承的故障状态(即轴承故障类型)。
其中，作为本发明一种较佳的具体实施例，故障诊断模型单元4将提取到的轴承的故障特征指标(向量)输入至训练好的基于梯度提升树(该算法为现有技术中的已有算法，在此不再赘述)的轴承故障诊断模型中进行模式识别，诊断出轴承的故障状态。
当装置处于故障诊断状态时，残差信号获取单元2采用全相位时移相位差校正法(该算法为现有技术中的已有算法，在此不再赘述)从待诊断的电机电流实时信号中获得电机电流实时信号的基频fe和谐波n*fe，n＝2、3、4、...、50的幅值、初始相位和频率，进而获得基频和谐波信号，再从原始的电机电流实时信号中减去基频和谐波信号，从而获得残差信号。
如附图2所示，当装置处于故障诊断状态时，故障特征提取单元3首 先计算轴承故障特征频率，再计算轴承故障特征频率序列。对残差信号进行FFT变换(Fast Fourier Transformation，快速傅里叶变换的简称)获得残差信号的频谱，再根据残差信号的频谱获得对应的轴承故障特征频率序列幅值。计算残差信号的最小值I_min和有效值I_rms，并根据轴承故障特征频率序列幅值I_f(n)，计算轴承故障特征频率序列幅值的最大值I_f_max、峰峰值I_f_peark和峰值系数I_f_crest。
故障诊断模型单元4，利用训练好的轴承故障诊断模型对故障特征提取单元3提取到的包括轴承故障特征频率序列幅值I_f(n)、轴承故障特征频率序列幅值的最大值I_f_max、峰峰值I_f_peark和峰值系数I_f_crest，以及残差信号的最小值I_min和有效值I_rms在内的轴承的故障特征指标(向量)进行模式识别，诊断出该轴承的故障状态。
当装置处于故障模型训练状态时：
电流信号获取单元1，获取在不同轴承故障状态(对应于不同的轴承故障类型)下的多组电机电流历史信号，作为模型的训练样本数据集；
残差信号获取单元2，对电流信号获取单元1获取的训练样本数据集进行谐波信号提取，从原始的电机电流历史信号中消除基频和谐波信号，获得残差信号；
故障特征提取单元3，对残差信号获取单元2得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
故障诊断模型单元4，基于故障特征提取单元3提取到的轴承的故障 特征指标(向量)并结合轴承故障类型进行训练，得到轴承故障诊断模型。
其中，作为本发明一种较佳的具体实施例，故障诊断模型单元4将提取到的轴承的故障特征指标(向量)及轴承故障类型输入至梯度提升树分类器(该算法为现有技术中的已有算法，在此不再赘述)中进行训练，得到轴承故障诊断模型。
当装置处于故障模型训练状态时，残差信号获取单元2采用全相位时移相位差校正法(该算法为现有技术中的已有算法，在此不再赘述)从原始的电机电流历史信号中获得电机电流历史信号的基频fe和谐波n*fe，n＝2、3、4、...、50的幅值、初始相位和频率，进而获得基频和谐波信号，再从原始的电机电流历史信号中减去基频和谐波信号，从而获得残差信号。
当装置处于故障模型训练状态时，故障特征提取单元3首先计算轴承故障特征频率，再计算轴承故障特征频率序列。轴承故障特征频率序列f(k)进一步为|fe±kfv|,k＝1,2,3,...,n，其中，n可取3～5次，fe为残差信号的电流基频频率，fv为轴承故障特征频率。对残差信号进行FFT变换获得残差信号的频谱，再根据残差信号的频谱获得对应的轴承故障特征频率序列幅值(即找到残差信号在各个频率点f(k)处的幅值，频率点f(k)与残差信号频谱中的某一频率相同)。计算残差信号的最小值I_min和有效值I_rms，并根据轴承故障特征频率序列幅值I_f(n)，计算轴承故障特征频率序列幅值的最大值I_f_max、峰峰值I_f_peark和峰值系数I_f_crest。
故障诊断模型单元4，基于故障特征提取单元3在不同轴承故障状态下提取到的包括轴承故障特征频率序列幅值I_f(n)、轴承故障特征频率序 列幅值的最大值I_f_max、峰峰值I_f_peark和峰值系数I_f_crest，以及残差信号的最小值I_min和有效值I_rms在内的轴承的故障特征指标(向量)，并结合轴承故障类型进行训练，得到轴承故障诊断模型。
故障特征提取单元3进一步根据以下公式计算残差信号的最小值I_min：
I_min＝min(I(t))
式中，I(t)为残差信号的时间序列，其中t＝1～N，N为信号的采样点数。
故障特征提取单元3进一步根据以下公式计算残差信号的有效值I_rms：
Figure PCTCN2019104202-appb-000006
故障特征提取单元3进一步根据以下公式计算轴承故障特征频率序列幅值的最大值：
I_f_max＝max(I_f(n))
式中，I_f(n)为轴承故障特征频率序列幅值。
故障特征提取单元3进一步根据以下公式计算轴承故障特征频率序列幅值的峰峰值：
I_f_peark＝I_f_max-I_min
故障特征提取单元3进一步根据以下公式计算轴承故障特征频率序列幅值的峰值系数：
I_f_crest＝I_f_max/I_rms
轴承的故障状态类型进一步包括轴承保持架故障、轴承外圈故障、轴承内圈故障和轴承滚动体故障。
故障特征提取单元3进一步根据以下公式计算轴承保持架故障特征频率：
Figure PCTCN2019104202-appb-000007
故障特征提取单元3进一步根据以下公式计算轴承外圈故障特征频率：
Figure PCTCN2019104202-appb-000008
故障特征提取单元3进一步根据以下公式计算轴承内圈故障特征频率：
Figure PCTCN2019104202-appb-000009
故障特征提取单元3进一步根据以下公式计算轴承滚动体故障特征频率：
Figure PCTCN2019104202-appb-000010
其中，fr为电机旋转频率，D B为轴承中滚动体的直径，D P为轴承节的圆直径，N B为轴承中滚动体的个数，θ为接触角。
本发明实施例1描述的电机轴承故障诊断装置的技术方案中，故障诊断模型单元4采用梯度提升树智能算法进行轴承故障诊断模型训练和模式 识别，也可以采用神经网络、随机森林和支持向量机等人工智能算法替代处理。实施例1描述的电机轴承故障诊断装置可很好地应用于机车牵引电机轴承故障的离线或在线健康监测与故障预警中。
实施例2
如附图3至附图5所示，一种基于实施例1所述装置的轴承故障诊断方法的实施例，具体包括以下步骤：
A)模型训练阶段
S101)获取在不同轴承故障状态下的多组电机电流历史信号，作为模型的训练样本数据集；
S102)对训练样本数据集进行谐波信号提取，从原始的电机电流历史信号中消除基频和谐波信号，获得残差信号；
S103)对步骤S102)得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
S104)基于步骤S103)提取到的轴承的故障特征指标(向量)并结合轴承故障类型进行训练，得到轴承故障诊断模型；
B)故障诊断阶段
S201)对待诊断的电机电流实时信号进行谐波信号提取，从原始的电机电流实时信号中消除基频和谐波信号，获得残差信号；
S202)对步骤S201)得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
S203)将步骤S202)提取到的轴承的故障特征指标(向量)输入至步骤S104)训练好的轴承故障诊断模型进行模式识别，诊断出该轴承的故障 状态。
通过实施本发明具体实施例描述的电机轴承故障诊断装置的技术方案，能够产生如下技术效果：
(1)本发明具体实施例描述的电机轴承故障诊断装置，采用对电机电流信号进行谐波信号提取，从原始的电机电流历史信号中消除基频和谐波信号，获得残差信号，有效减少了由PWM供电引入的谐波信号对轴承故障特征信号的干扰，提高了电流信号的信噪比，进而提取的故障特征指标更加显著；
(2)本发明具体实施例描述的电机轴承故障诊断装置，利用残差信号提取的轴承故障特征指标的变化来反映牵引电机轴承健康状态的变化，该故障特征指标能够准确地表征轴承故障状态，采用全相位时移相位差校正法提取并去除电流信号中的谐波信号，进一步提升了故障特征指标的显著性；
(3)本发明具体实施例描述的电机轴承故障诊断装置，所建立的梯度提升树故障诊断模型可自动识别轴承故障状态，而不需要业务专家观察频谱图进行诊断，能够根据电机电流信号的自身特性自动地识别轴承故障与否，有效减少了人为主观因素的干扰，并提高了故障诊断的客观性和准确性。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。
以上所述，仅是本发明的较佳实施例而已，并非对本发明作任何形式 上的限制。虽然本发明已以较佳实施例揭示如上，然而并非用以限定本发明。任何熟悉本领域的技术人员，在不脱离本发明的精神实质和技术方案的情况下，都可利用上述揭示的方法和技术内容对本发明技术方案做出许多可能的变动和修饰，或修改为等同变化的等效实施例。因此，凡是未脱离本发明技术方案的内容，依据本发明的技术实质对以上实施例所做的任何简单修改、等同替换、等效变化及修饰，均仍属于本发明技术方案保护的范围。

Claims (10)

  1. 一种电机轴承故障诊断装置，其特征在于，包括：电流信号获取单元(1)、残差信号获取单元(2)、故障特征提取单元(3)和故障诊断模型单元(4)；当所述装置处于故障诊断状态时：
    电流信号获取单元(1)，获取电机电流实时信号；
    残差信号获取单元(2)，对所述电流信号获取单元(1)获取的电机电流实时信号进行谐波信号提取，从原始的电机电流实时信号中消除基频和谐波信号，获得残差信号；
    故障特征提取单元(3)，对所述残差信号获取单元(2)得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
    故障诊断模型单元(4)，利用训练好的轴承故障诊断模型对所述故障特征提取单元(3)提取到的轴承的故障特征指标进行模式识别，诊断出该轴承的故障状态。
  2. 根据权利要求1所述的电机轴承故障诊断装置，其特征在于：当所述装置处于故障诊断状态时，所述故障特征提取单元(3)首先计算轴承故障特征频率，再计算轴承故障特征频率序列；对残差信号进行FFT变换获得残差信号的频谱，再根据残差信号的频谱获得对应的轴承故障特征频率序列幅值；计算残差信号的最小值(I_min)和有效值(I_rms)，并根据轴承故障特征频率序列幅值(I_f(n))，计算轴承故障特征频率序列幅值的最大值(I_f_max)、峰峰值(I_f_peark)和峰值系数(I_f_crest)。
  3. 根据权利要求1或2所述的电机轴承故障诊断装置，其特征在于， 当所述装置处于故障模型训练状态时：
    电流信号获取单元(1)，获取在不同轴承故障状态下的多组电机电流历史信号，作为模型的训练样本数据集；
    残差信号获取单元(2)，对所述电流信号获取单元(1)获取的训练样本数据集进行谐波信号提取，从原始的电机电流历史信号中消除基频和谐波信号，获得残差信号；
    故障特征提取单元(3)，对所述残差信号获取单元(2)得到的残差信号进行时域和频域分析，提取轴承的故障特征指标；
    故障诊断模型单元(4)，基于所述故障特征提取单元(3)提取到的轴承的故障特征指标并结合轴承故障类型进行训练，得到轴承故障诊断模型。
  4. 根据权利要求3所述的电机轴承故障诊断装置，其特征在于：
    当所述装置处于故障模型训练状态时，所述残差信号获取单元(2)从原始的电机电流历史信号中获得电机电流历史信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号，再从原始的电机电流历史信号中减去基频和谐波信号，从而获得残差信号；
    当所述装置处于故障诊断状态时，所述残差信号获取单元(2)从待诊断的电机电流实时信号中获得电机电流实时信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号，再从原始的电机电流实时信号中减去基频和谐波信号，从而获得残差信号。
  5. 根据权利要求4所述的电机轴承故障诊断装置，其特征在于：当所 述装置处于故障模型训练状态时，所述故障特征提取单元(3)首先计算轴承故障特征频率，再计算轴承故障特征频率序列；对残差信号进行FFT变换获得残差信号的频谱，再根据残差信号的频谱获得对应的轴承故障特征频率序列幅值；计算残差信号的最小值(I_min)和有效值(I_rms)，并根据轴承故障特征频率序列幅值(I_f(n))，计算轴承故障特征频率序列幅值的最大值(I_f_max)、峰峰值(I_f_peark)和峰值系数(I_f_crest)。
  6. 根据权利要求4或5所述的电机轴承故障诊断装置，其特征在于：当所述装置处于故障模型训练状态时，所述残差信号获取单元(2)采用全相位时移相位差校正法获得电机电流历史信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号；当所述装置处于故障诊断状态时，所述残差信号获取单元(2)采用全相位时移相位差校正法获得电机电流实时信号的基频和谐波的幅值、初始相位和频率，进而获得基频和谐波信号。
  7. 根据权利要求6所述的电机轴承故障诊断装置，其特征在于：当所述装置处于故障模型训练状态时，所述故障诊断模型单元(4)利用梯度提升树分类器对提取到的轴承的故障特征指标及轴承故障类型进行训练，得到轴承故障诊断模型；当所述装置处于故障诊断状态时，所述故障诊断模型单元(4)利用训练好的基于梯度提升树的轴承故障诊断模型对提取到的轴承的故障特征指标进行模式识别，诊断出轴承的故障状态。
  8. 根据权利要求1、2、4、5或7任一项所述的电机轴承故障诊断装置，其特征在于：所述轴承故障特征频率序列f(k)为|fe±kfv|,k＝1,2,3，其中，fe为残差信号的电流基频频率，fv为轴承故障特征频率。
  9. 根据权利要求8所述的电机轴承故障诊断装置，其特征在于，所述故障特征提取单元(3)根据以下公式计算所述残差信号的最小值(I_min)：
    I_min＝min(I(t))
    式中，I(t)为残差信号的时间序列，其中t＝1～N，N为信号的采样点数；
    根据以下公式计算所述残差信号的有效值(I_rms)：
    Figure PCTCN2019104202-appb-100001
    根据以下公式计算所述轴承故障特征频率序列幅值的最大值：
    I_f_max＝max(I_f(n))
    式中，I_f(n)为轴承故障特征频率序列幅值；
    根据以下公式计算所述轴承故障特征频率序列幅值的峰峰值：
    I_f_peark＝I_f_max-I_min
    根据以下公式计算所述轴承故障特征频率序列幅值的峰值系数：
    I_f_crest＝I_f_max/I_rms。
  10. 根据权利要求1、2、4、5、7或9任一项所述的电机轴承故障诊断装置，其特征在于：所述轴承的故障状态类型包括轴承保持架故障、轴承外圈故障、轴承内圈故障和轴承滚动体故障；
    所述故障特征提取单元(3)根据以下公式计算所述轴承保持架故障特征频率：
    Figure PCTCN2019104202-appb-100002
    所述故障特征提取单元(3)根据以下公式计算所述轴承外圈故障特征频率：
    Figure PCTCN2019104202-appb-100003
    所述故障特征提取单元(3)根据以下公式计算所述轴承内圈故障特征频率：
    Figure PCTCN2019104202-appb-100004
    所述故障特征提取单元(3)根据以下公式计算所述轴承滚动体故障特征频率：
    Figure PCTCN2019104202-appb-100005
    其中，fr为电机旋转频率，D B为轴承中滚动体的直径，D P为轴承节的圆直径，N B为轴承中滚动体的个数，θ为接触角。
PCT/CN2019/104202 2018-10-15 2019-09-03 一种电机轴承故障诊断装置 WO2020078132A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US17/261,886 US11898932B2 (en) 2018-10-15 2019-09-03 Motor bearing failure diagnosis device
EP19872841.2A EP3816604B1 (en) 2018-10-15 2019-09-03 Motor bearing failure diagnosis device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811197298.4 2018-10-15
CN201811197298.4A CN109061474A (zh) 2018-10-15 2018-10-15 一种电机轴承故障诊断装置

Publications (1)

Publication Number Publication Date
WO2020078132A1 true WO2020078132A1 (zh) 2020-04-23

Family

ID=64764815

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/104202 WO2020078132A1 (zh) 2018-10-15 2019-09-03 一种电机轴承故障诊断装置

Country Status (4)

Country Link
US (1) US11898932B2 (zh)
EP (1) EP3816604B1 (zh)
CN (1) CN109061474A (zh)
WO (1) WO2020078132A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111650514A (zh) * 2020-06-15 2020-09-11 珠海万力达电气自动化有限公司 一种异步电动机典型故障的多参数联合诊断方法
CN112052796A (zh) * 2020-09-07 2020-12-08 电子科技大学 一种基于深度学习的永磁同步电机故障诊断方法
CN113075545A (zh) * 2021-03-22 2021-07-06 西安工业大学 一种电机故障诊断结果的表现方法
CN115616403A (zh) * 2022-12-13 2023-01-17 湖南大学 永磁同步电机轴承故障初期诊断方法、装置及存储介质
CN116106742A (zh) * 2023-04-07 2023-05-12 国家石油天然气管网集团有限公司 一种基于dq变换的电机轴承故障诊断方法及系统

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109061474A (zh) 2018-10-15 2018-12-21 株洲中车时代电气股份有限公司 一种电机轴承故障诊断装置
US11506569B2 (en) * 2019-01-15 2022-11-22 Computational Systems, Inc. Bearing and fault frequency identification from vibration spectral plots
US11099101B2 (en) * 2019-05-03 2021-08-24 Mitsubishi Electric Research Laboratories, Inc. Method for estimating bearing fault severity for induction motors
CN110823568A (zh) * 2019-09-29 2020-02-21 深圳精匠云创科技有限公司 轴承故障检测装置及检测方法
CN111220379B (zh) * 2020-04-23 2020-07-31 湖南中车时代通信信号有限公司 一种牵引电机传动系统故障诊断方法及装置
CN111208427B (zh) * 2020-04-23 2020-07-24 湖南中车时代通信信号有限公司 一种牵引电机故障诊断方法及装置
CN111654141A (zh) * 2020-06-03 2020-09-11 宁波大学 一种电磁感应式轴承转速测量装置及测量方法
CN111680665B (zh) * 2020-06-28 2023-06-20 湖南大学 基于数据驱动的采用电流信号的电机机械故障诊断方法
EP3957970A1 (en) * 2020-08-17 2022-02-23 ABB Schweiz AG Method of detecting a bearing fault
CN112071210A (zh) * 2020-09-18 2020-12-11 广州花筑科技有限公司 一种地铁口信息提示牌
CN112320520B (zh) * 2020-11-09 2022-12-06 浙江新再灵科技股份有限公司 基于残差分析的电梯异常振动检测方法
CN112611563B (zh) * 2020-12-01 2023-07-14 上海明略人工智能(集团)有限公司 一种目标故障信息的确定方法和装置
GB2607975A (en) * 2021-06-17 2022-12-21 Eaton Intelligent Power Ltd Method and apparatus for motor bearing fault detection using sensed current
EP4356096A1 (en) * 2021-06-17 2024-04-24 Eaton Intelligent Power Limited Method and apparatus for motor bearing fault detection using sensed current
KR102601072B1 (ko) * 2021-10-21 2023-11-09 서울대학교산학협력단 모터 고장 진단을 위한 이미지 생성 방법 및 장치, 상기 이미지를 이용한 모터 고장 진단 방법 및 장치
CN114112396B (zh) * 2021-10-29 2023-09-26 江苏大学 一种采用电流信号分析的转速波动准平稳工况下轴承故障诊断方法
CN114114008B (zh) * 2021-11-10 2024-01-05 江苏澳洋医药物流有限公司 一种智能医药货架的输送带电机故障诊断装置
CN114925735A (zh) * 2022-07-21 2022-08-19 深圳市世坤科技实业有限公司 新能源汽车电机的运行状态评估方法
CN115062678B (zh) * 2022-08-19 2022-11-04 山东能源数智云科技有限公司 设备故障检测模型的训练方法、故障检测方法及装置
CN115270896B (zh) * 2022-09-28 2023-04-07 西华大学 一种用于识别航空发动机主轴承松动故障的智能诊断方法
CN116304905B (zh) * 2023-02-03 2024-01-23 中国人民解放军92942部队 一种在多负载工况下的永磁同步电机退磁故障诊断方法
CN116520143B (zh) * 2023-07-03 2023-09-12 利维智能（深圳）有限公司 基于声纹数据的旋转设备监测方法、装置、设备及介质
CN116756675B (zh) * 2023-08-14 2023-11-07 湘江实验室 一种基于脉冲神经网络的轴承故障诊断分类方法
CN116990744B (zh) * 2023-09-25 2023-12-05 北京志翔科技股份有限公司 一种电能表检测方法、装置、设备及介质

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015129523A (ja) * 2014-01-06 2015-07-16 日本精工株式会社 転がり軸受の使用方法
CN105929331A (zh) * 2016-04-28 2016-09-07 上海电机学院 一种双馈风力发电机定转子故障分析诊断装置及方法
CN106680716A (zh) 2017-01-20 2017-05-17 浙江大学 一种基于无位置传感器的永磁电机轴承故障诊断的方法
CN106769041A (zh) 2016-12-23 2017-05-31 安徽大学 一种变转速工况下的永磁同步电机轴承在线故障诊断装置及方法
CN106989923A (zh) 2017-03-28 2017-07-28 南京航空航天大学 基于定子电流小波包分析的永磁机轴承点蚀故障检测方法
CN108303249A (zh) * 2017-01-12 2018-07-20 天津工业大学 一种用于齿轮箱电信号残差信号的提取方法
CN109061474A (zh) * 2018-10-15 2018-12-21 株洲中车时代电气股份有限公司 一种电机轴承故障诊断装置
CN109238698A (zh) * 2018-10-15 2019-01-18 株洲中车时代电气股份有限公司 一种基于电流信号的电机轴承故障诊断方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6199023B1 (en) * 1998-11-16 2001-03-06 Geneal Electric Company System for removing spurious signatures in motor current signature analysis
US7117125B2 (en) * 2003-06-18 2006-10-03 Eaton Corporation System and method for proactive motor wellness diagnosis based on potential mechanical faults
CN100543440C (zh) * 2007-03-28 2009-09-23 华北电力大学 异步电动机轴承故障检测方法
US7847580B2 (en) * 2007-06-04 2010-12-07 Eaton Corporation System and method for motor fault detection using stator current noise cancellation
US8994359B2 (en) 2011-08-29 2015-03-31 General Electric Company Fault detection based on current signature analysis for a generator
US11105712B2 (en) * 2018-04-06 2021-08-31 Raytheon Technologies Corporation Integrated vibe/ODM fusion and trending analysis for prognostic health management of engine bearing

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015129523A (ja) * 2014-01-06 2015-07-16 日本精工株式会社 転がり軸受の使用方法
CN105929331A (zh) * 2016-04-28 2016-09-07 上海电机学院 一种双馈风力发电机定转子故障分析诊断装置及方法
CN106769041A (zh) 2016-12-23 2017-05-31 安徽大学 一种变转速工况下的永磁同步电机轴承在线故障诊断装置及方法
CN108303249A (zh) * 2017-01-12 2018-07-20 天津工业大学 一种用于齿轮箱电信号残差信号的提取方法
CN106680716A (zh) 2017-01-20 2017-05-17 浙江大学 一种基于无位置传感器的永磁电机轴承故障诊断的方法
CN106989923A (zh) 2017-03-28 2017-07-28 南京航空航天大学 基于定子电流小波包分析的永磁机轴承点蚀故障检测方法
CN109061474A (zh) * 2018-10-15 2018-12-21 株洲中车时代电气股份有限公司 一种电机轴承故障诊断装置
CN109238698A (zh) * 2018-10-15 2019-01-18 株洲中车时代电气股份有限公司 一种基于电流信号的电机轴承故障诊断方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3816604A4

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111650514A (zh) * 2020-06-15 2020-09-11 珠海万力达电气自动化有限公司 一种异步电动机典型故障的多参数联合诊断方法
CN111650514B (zh) * 2020-06-15 2022-09-09 珠海万力达电气自动化有限公司 一种异步电动机典型故障的多参数联合诊断方法
CN112052796A (zh) * 2020-09-07 2020-12-08 电子科技大学 一种基于深度学习的永磁同步电机故障诊断方法
CN113075545A (zh) * 2021-03-22 2021-07-06 西安工业大学 一种电机故障诊断结果的表现方法
CN113075545B (zh) * 2021-03-22 2024-02-23 西安工业大学 一种电机故障诊断结果的表现方法
CN115616403A (zh) * 2022-12-13 2023-01-17 湖南大学 永磁同步电机轴承故障初期诊断方法、装置及存储介质
CN116106742A (zh) * 2023-04-07 2023-05-12 国家石油天然气管网集团有限公司 一种基于dq变换的电机轴承故障诊断方法及系统

Also Published As

Publication number Publication date
EP3816604A4 (en) 2022-04-06
EP3816604A1 (en) 2021-05-05
US11898932B2 (en) 2024-02-13
US20210286995A1 (en) 2021-09-16
CN109061474A (zh) 2018-12-21
EP3816604B1 (en) 2024-05-01

Similar Documents

Publication Publication Date Title
WO2020078132A1 (zh) 一种电机轴承故障诊断装置
CN109238698A (zh) 一种基于电流信号的电机轴承故障诊断方法
Li et al. Application of bandwidth EMD and adaptive multiscale morphology analysis for incipient fault diagnosis of rolling bearings
Yang et al. Cost-effective condition monitoring for wind turbines
CN107356432B (zh) 基于频域窗经验小波共振解调的滚动轴承故障诊断方法
Zhang et al. A novel detection method of motor broken rotor bars based on wavelet ridge
WO2022261805A1 (zh) 一种柴油机齿轮箱故障诊断方法
Choqueuse et al. Current frequency spectral subtraction and its contribution to induction machines’ bearings condition monitoring
Kia et al. A comparative study of acoustic, vibration and stator current signatures for gear tooth fault diagnosis
KR101169797B1 (ko) 3상 유도전동기의 고정자 권선 고장 진단시스템
CN103698699A (zh) 一种基于模型的异步电动机故障监测诊断方法
CN110163190B (zh) 一种滚动轴承故障诊断方法及装置
Wei et al. Time-varying envelope filtering for exhibiting space bearing cage fault features
Amirat et al. Condition monitoring of wind turbines based on amplitude demodulation
CN111881736A (zh) 一种基于带宽傅立叶分解的滚动轴承早期故障诊断方法
Duan et al. Adaptive morphological analysis method and its application for bearing fault diagnosis
WO2020057066A1 (zh) 基于增强调制双谱分析的滚动轴承故障诊断方法
Guo et al. Automatic bearing fault diagnosis of permanent magnet synchronous generators in wind turbines subjected to noise interference
CN108278184B (zh) 基于经验模态分解的风电机组叶轮不平衡监测方法
Liu et al. An online bearing fault diagnosis technique via improved demodulation spectrum analysis under variable speed conditions
Kumar et al. A combined approach for weak fault signature extraction of rolling element bearing using Hilbert envelop and zero frequency resonator
Zhang et al. Improved local cepstrum and its applications for gearbox and rolling bearing fault detection
CN104111391A (zh) 一种基于三相瞬时功率的电主轴故障监测诊断方法
CN102495264B (zh) 基于信息熵的电压跌落持续时间检测方法
Abolghasemi et al. Recursive singular spectrum analysis for induction machines unbalanced rotor fault diagnosis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19872841

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2019872841

Country of ref document: EP

Effective date: 20210129

NENP Non-entry into the national phase

Ref country code: DE