WO2024012199A1 - 多维度数据融合电机故障预检装置、系统及其预检方法 - Google Patents

多维度数据融合电机故障预检装置、系统及其预检方法 Download PDF

Info

Publication number
WO2024012199A1
WO2024012199A1 PCT/CN2023/103024 CN2023103024W WO2024012199A1 WO 2024012199 A1 WO2024012199 A1 WO 2024012199A1 CN 2023103024 W CN2023103024 W CN 2023103024W WO 2024012199 A1 WO2024012199 A1 WO 2024012199A1
Authority
WO
WIPO (PCT)
Prior art keywords
motor
under test
fault
data
dimensional data
Prior art date
Application number
PCT/CN2023/103024
Other languages
English (en)
French (fr)
Inventor
黄峰
雷欢
傅阳
杜姗
吴晓峰
Original Assignee
浙江联宜电机有限公司
浙江科技学院
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 浙江联宜电机有限公司, 浙江科技学院 filed Critical 浙江联宜电机有限公司
Publication of WO2024012199A1 publication Critical patent/WO2024012199A1/zh

Links

Classifications

    • 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
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D21/00—Measuring or testing not otherwise provided for
    • G01D21/02—Measuring two or more variables by means not covered by a single other subclass
    • 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
    • G01R31/343—Testing dynamo-electric machines in operation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning

Definitions

  • the present invention relates to a motor fault prediction and diagnosis, and in particular to a multi-dimensional data fusion motor fault pre-checking device and system used in the prediction and diagnosis of motor faults and a pre-checking method thereof.
  • motors are widely used in different fields and can be seen everywhere in our daily production and life.
  • their internal structures and operating technology types are becoming more and more complex, which greatly increases the probability of equipment failure.
  • the resulting economic losses may be unpredictable. At the least, it may cause equipment operation interruption and productivity interruption and loss, and at worst, it may pose a threat to human lives and corporate assets.
  • the traditional maintenance strategy adopted after a motor failure is to have professionals make on-site diagnosis and maintenance under limited maintenance resources. This is also difficult to solve immediately as soon as the motor fails. As the downtime increases, If it is stretched again and again, the losses caused by motor failure will be greater.
  • the fault characteristic information that can be extracted by using a single sensor is limited, making the fault diagnosis results unreliable.
  • data fusion technology that uses data measured by multiple different types of sensors becomes particularly important. It can extract more accurate and comprehensive fault characteristic information and improve the accuracy and reliability of motor fault pre-detection.
  • the invention provides a method for predicting and monitoring motor operation faults in order to solve the current situation that after the existing motor fails in operation, it is easy to cause production stoppage and other varying degrees of losses due to the inability to obtain timely and effective repair and maintenance of the motor fault. , timely monitoring and pre-inspection to obtain different hidden fault phenomena in motor operation, improve the accuracy and reliability of motor failure pre-inspection, and bring safety guarantees for timely inspection and maintenance of motor faults. Multi-dimensional data fusion motor fault pre-inspection devices, systems and their prediction inspection method.
  • a multi-dimensional data fusion motor fault pre-detection device including a tested motor and its motor drive controller, which is characterized in that: it also includes a loading motor and its motor drive controller, and four-dimensional sensing modules that collect raw data from four dimensions when the motor under test is working.
  • the four-dimensional sensing modules are electrical dimension sensing module, mechanical dimension sensing module, acoustic dimension sensing module and Temperature dimension sensing module, where the electrical sensing module is used to collect voltage, current and power data when the motor is working, the mechanical sensing module is used to collect vibration, torque and speed data from the mechanical dimension when the motor is working, and the acoustic dimension
  • the sensing module is used to collect the noise data emitted by the motor when it is working from the acoustic dimension
  • the temperature dimension sensing module is used to collect the surface temperature change data when the motor is working from the temperature dimension; the motor under test and the loading motor are both located in the same workstation.
  • the four-dimensional sensing modules are electrically connected to the data acquisition card.
  • the multi-dimensional sensing module is used to collect multi-dimensional data information when the motor is working, which can predict and monitor motor operating faults. Timely monitoring and pre-inspection can obtain different hidden fault phenomena in motor operation, improve the accuracy and reliability of motor fault pre-inspection, and provide Motor faults can be repaired and maintained in time to ensure safety.
  • the mechanical dimension sensing module uses a torque sensor and a vibration sensor, where the torque sensor is located between the two couplings connecting the motor under test and the loading motor; the vibration sensor uses a magnetic connection seat, through The magnetic connection base is magnetically adsorbed and fixed on the upper end of the motor casing under test.
  • the torque sensor is located between the two couplings connecting the motor under test and the loading motor
  • the vibration sensor uses a magnetic connection seat, through
  • the magnetic connection base is magnetically adsorbed and fixed on the upper end of the motor casing under test.
  • the motor under test and the loading motor both adopt a shaft-out design structure.
  • the torque sensor configured in the mechanical dimension sensing module also adopts a shaft-out design structure.
  • the shaft-out part of the motor under test, the torque sensor and the loading motor Connect in sequence. Improve the working coaxial requirements of the motor under test, torque sensor and loading motor.
  • the workbench adopts a T-slot workbench structure.
  • the T-slot workbench is equipped with a motor mounting bracket under test, a loading motor mounting bracket, a torque sensor mounting bracket and an acoustic dimension sensor mounting bracket.
  • the motor under test passes through The tested motor mounting bracket is connected and fixed on the T-slot workbench, the loading motor is connected and fixed on the T-slot workbench through the loading motor mounting bracket, and the torque sensor is connected and fixed on the T-slot workbench through the torque sensor mounting bracket.
  • the acoustic dimension sensor is connected and fixed on the T-slot workbench through the acoustic dimension sensor mounting bracket, while maintaining the output shaft part of the motor under test, the torque sensor and the loading motor to meet the coaxiality requirements.
  • Improve the convenience, flexibility and reliability of the installation, use, disassembly and maintenance of the motor mounting bracket, loading motor mounting bracket, torque sensor mounting bracket and acoustic dimension sensor mounting bracket improve the convenience, reliability and effectiveness of installation and positioning, and improve the reliability of coaxiality requirements.
  • the temperature dimension sensing module adopts a temperature sensor, and the temperature sensor adopts a magnet-type patch temperature sensor, which is directly adsorbed on the motor casing above the bearing of the motor under test.
  • the temperature sensor adopts a magnet-type patch temperature sensor, which is directly adsorbed on the motor casing above the bearing of the motor under test.
  • the electrical dimension sensing module uses a power meter that collects the voltage, current and power signals output by the motor under test.
  • the power meter receives the power signal from the drive motor controller and transmits it to the motor under test, and at the same time outputs the power signal being measured. Measure the voltage, current and power signals of the motor when it is working. Improve the convenience, reliability and effectiveness of the working signal output collection of the motor under test.
  • Another object of the present invention is to provide a multi-dimensional data fusion motor fault pre-detection system, which is characterized in that: the multi-dimensional data fusion motor fault pre-detection device described in one of the above technical solutions is adopted, in which the motor under test and Its motor drive controller forms the drive system in the pre-inspection system, and the loading motor and its motor drive controller form the loading system in the pre-inspection system.
  • the output ends of the motor under test and the loading motor are equipped with couplings, and the two couplings
  • a torque sensor is installed between the shafts.
  • the two couplings form a transmission system.
  • the transmission system is connected between the motor under test and the loading motor.
  • the four-dimensional sensing module and the data acquisition card form a data acquisition system.
  • the system is electrically connected to the PC terminal of the pre-inspection computer.
  • the multi-dimensional sensing module is used to collect multi-dimensional data information when the motor is working, which can predict and monitor motor operating faults. Timely monitoring and pre-inspection can obtain different hidden fault phenomena in motor operation, improve the accuracy and reliability of motor fault pre-inspection, and provide Motor faults can be repaired and maintained in time to ensure safety.
  • the coupling adopts an elastic element flexible coupling. It improves the relative displacement compensation of the motor under test, the loading motor and the two couplings, compensates for the coaxiality error of the torque sensor, the motor under test and the load motor, and has the ability to buffer and absorb shock when the motor under test is running at high speed. Especially necessary.
  • Another object of the present invention is a multi-dimensional data fusion motor fault pre-checking method, which is characterized by: including the following pre-checking steps
  • the multi-dimensional data fusion motor fault pre-detection device described in one of the above technical solutions collects raw data from four dimensions when the tested motor is working;
  • the multi-dimensional data fusion motor fault pre-detection system described in one of the above technical solutions performs data cleaning on the original data of the tested motor in four dimensions collected in the above S1 step.
  • the data cleaning includes missing data and Data cleaning processing of noisy data;
  • a classification neural network is used to diagnose the fault of the motor under test.
  • the multi-dimensional sensing module is used to collect multi-dimensional data information when the motor is working, which can predict and monitor motor operating faults. Timely monitoring and pre-inspection can obtain different hidden fault phenomena in motor operation, improve the accuracy and reliability of motor fault pre-inspection, and provide Motor faults can be repaired and maintained in time to ensure safety.
  • Two neural networks are used, one to perform data fusion on multi-dimensional information of the motor status to extract fault features, and the other to perform fault diagnosis on the extracted fault feature information. This avoids the complex modeling process and solves the problem of limited fault feature information extracted by a single sensor. This makes the fault diagnosis results unreliable.
  • the multi-dimensional data fusion technology is to build a neural network model of Transformer+SAE model architecture, and use the multi-dimensional data after data preprocessing by the data acquisition board in the data acquisition system as the input sample Input into the trained Transformer+SAE model architecture, perform multi-dimensional data fusion to extract important fault feature information;
  • the classification neural network uses the Softmax multi-classifier as the classification network, and the Transformer+SAE model architecture
  • the output important fault characteristic information is input into the trained classification neural network as an input sample.
  • the output result of the model is the probability result that the motor under test belongs to various faults. The largest probability among all output probabilities is the final motor fault diagnosis. result.
  • Two neural networks are used, one to perform data fusion on multi-dimensional information of the motor status to extract fault features, and the other to perform fault diagnosis on the extracted fault feature information. This avoids the complex modeling process and solves the problem of limited fault feature information extracted by a single sensor. This makes the fault diagnosis results unreliable.
  • the multi-dimensional sensing module is used to collect multi-dimensional data information when the motor is working, which can predict and monitor motor operating faults. Timely monitoring and pre-inspection can obtain different hidden fault phenomena in motor operation, improve the accuracy and reliability of motor fault pre-inspection, and provide Motor faults can be repaired and maintained in time to ensure safety.
  • Two neural networks are used, one to perform data fusion on multi-dimensional information of the motor status to extract fault features, and the other to perform fault diagnosis on the extracted fault feature information. This avoids the complex modeling process and solves the problem of limited fault feature information extracted by a single sensor. This makes the fault diagnosis results unreliable.
  • the present invention adopts the deep neural network model architecture of Transformer+SAE.
  • Transformer as an efficient feature extractor, has strong parallel computing capabilities and faster model training efficiency. It not only solves the problem of short-term memory of the RNN model, but also makes up for it. It overcomes the limitations of CNN's ability to capture long-distance features.
  • SAE is an unsupervised algorithm that can automatically learn features from unlabeled data and can provide better feature descriptions than the original data, thereby automatically completing the dimensionality reduction process. Through the combination of the two, it is possible to reduce the amount of calculation and improve the efficiency of data fusion, while also increasing the reliability of motor fault diagnosis results.
  • Figure 1 is a schematic structural diagram of the principle structure of the multi-dimensional data fusion motor fault pre-checking device, system and pre-checking method of the present invention.
  • Figure 2 is a schematic structural diagram of the multi-dimensional data fusion motor fault pre-detection device of the present invention.
  • Figure 3 is a schematic structural diagram of the principle structure of the multi-dimensional data fusion motor fault pre-inspection system of the present invention.
  • Figure 4 is a schematic diagram of the electrical structure of the multi-dimensional data fusion motor fault pre-detection system of the present invention.
  • Figure 5 is an overall structural diagram of the motor fault diagnosis network of the multi-dimensional data fusion technology of Zhang Zhong's multi-dimensional data fusion motor fault pre-detection method according to the present invention.
  • Figure 6 is a schematic diagram of the Transformer+SAE model architecture of the multi-dimensional data fusion motor fault pre-detection method of the present invention.
  • Figure 7 is a schematic diagram of the network structure of Zhang Zhong's Softmax multi-classifier in the multi-dimensional data fusion motor fault pre-detection method of the present invention.
  • a multi-dimensional data fusion motor fault pre-detection device includes a motor 10 under test and its motor drive controller, a loading motor 20 and its motor drive controller, and a slave motor.
  • the four-dimensional sensing module (multi-dimensional data) collects the original data of the motor under test in four dimensions.
  • the four-dimensional sensing modules are the electrical dimension sensing module, the mechanical dimension sensing module, the acoustic dimension sensing module and the temperature. Dimensional sensing module.
  • the electrical sensing module is used to collect the voltage, current and power data of the motor when it is working from the electrical dimension.
  • the mechanical sensing module is used to collect the vibration, torque and speed data of the motor when it is working from the mechanical dimension.
  • the acoustic dimension sensing module is used to collect the noise data emitted by the motor when it is working from the acoustic dimension
  • the temperature dimension sensing module is used to collect the surface temperature change data when the motor is working from the temperature dimension; both the motor under test and the loading motor are equipped with
  • the four-dimensional sensing modules are electrically connected to the data acquisition card.
  • the data acquisition board is used to receive information collected by the sensing modules of each dimension.
  • the processor on the data acquisition board can process the data, perform simple data cleaning and data fusion, and transmit fault characteristics to the pre-inspection computer PC 70 .
  • the mechanical dimension sensing module uses a torque sensor 14 and a vibration sensor 12.
  • the torque sensor 14 is installed and connected between the two couplings 15 that connect the motor under test 10 and the loading motor 20; the vibration sensor 12 uses a magnetic connection.
  • the base is magnetically adsorbed and fixed on the upper end of the housing of the motor 10 under test through the magnetic connection base.
  • the vibration sensor 12 can simultaneously collect vibration information in the horizontal direction and vibration information in the vertical direction.
  • Both the motor under test 10 and the loading motor 20 adopt an out-shaft design structure.
  • the torque sensor 14 configured in the mechanical dimension sensing module also adopts an out-shaft design structure.
  • the output ends of the motor under test 10 and the loading motor 20 are equipped with couplings.
  • the motor 10 under test, the torque sensor 14 and the output shaft part of the loading motor 20 are connected in sequence through the coupling 14 respectively.
  • the motor to be tested, the torque sensor, and the output shaft part of the loading motor can be connected through two couplings 14.
  • the workbench 80 adopts a T-slot workbench structure.
  • the T-slot workbench is equipped with a motor mounting bracket 16, a loading motor mounting bracket 21, a torque sensor mounting bracket and an acoustic dimension sensor mounting bracket 17.
  • the motor being tested 10 passes The motor mounting bracket 16 under test is connected and fixed on the T-slot workbench, the loading motor 20 is connected and fixed on the T-slot workbench through the loading motor mounting bracket 21, and the torque sensor 14 is connected and fixed on the T-slot workbench through the torque sensor mounting bracket 141.
  • the acoustic dimension sensor 14 is connected and fixed on the T-slot workbench through the acoustic dimension sensor mounting bracket, while maintaining the coaxiality of the output shaft parts of the motor under test 10, the torque sensor 14 and the loading motor 20.
  • the acoustic sensing module refers to the sound sensor 13 used to collect the sound signal of the motor under test when it is working.
  • the temperature dimension sensing module uses a temperature sensor 11.
  • the temperature sensor 11 uses a magnet-type patch temperature sensor, which is directly adsorbed on the motor shell above the bearing of the motor 10 under test, making it easy to install and disassemble.
  • the temperature dimension sensing module refers to the temperature sensor 11 used to collect the temperature signal WDXH of the motor under test when it is working.
  • the electrical dimension sensing module uses a power meter that collects the voltage, current and power signal YLGXH output by the motor under test. The power meter receives data from the driver. The power signal sent by the motor controller is transmitted to the motor under test, and the voltage, current and power signal YLGXH of the motor under test is output at the same time.
  • a multi-dimensional data fusion motor fault pre-checking system adopts the multi-dimensional data fusion motor failure pre-check device described in Embodiment 1, in which
  • the test motor and its motor drive controller constitute the drive system in the pre-inspection system.
  • the drive motor controller 40 is used to control the input power of the motor under test and input the control signal KZXH to the motor under test 10.
  • the drive motor controller 40 can send control Signals and power signals are given to the motor under test (or called the motor under test) to control the motor under test to rotate in the set direction and speed; the loading motor 20 and its motor drive controller form the loading system in the pre-inspection system, and the loading motor
  • the controller 50 is used to simulate the actual load condition of the motor under test.
  • the loading motor controller 50 transmits the power signal DLXH and the loading control signal JKZXH to the loading motor 20, and controls the loading motor to rotate according to the set direction and speed.
  • the purpose is to simulate the loading motor 20. Measure the actual load condition of the motor; the output ends of the tested motor 10 and the loading motor 20 are both equipped with a coupling 15, and a torque sensor is provided between the two couplings.
  • the two couplings form a transmission system.
  • the transmission system The connection is between the motor under test and the loading motor.
  • the four-dimensional sensing module and the data acquisition card form a data acquisition system.
  • the data acquisition system includes an electrical sensing module, a mechanical sensing module, an acoustic sensing module, and a temperature sensing module. and data acquisition board 60.
  • the electrical sensing module includes a power meter 30 for collecting the voltage, current, and power signals YLGXH output by the motor under test.
  • the power meter can receive the power signal from the drive motor controller and transmit it to the power signal under test.
  • the motor simultaneously outputs voltage, current, and power signals when the motor is working;
  • the mechanical sensing module includes a vibration sensor used to collect the vibration signal ZDXH of the motor under test and a torque sensor used to collect the torque and speed signal NJXH of the motor under test;
  • the acoustic sensor module includes a sound sensor used to collect the sound signal SYXH when the motor under test is working;
  • the temperature sensing module includes a temperature sensor used to collect the temperature signal when the motor under test is working;
  • the data acquisition board is mainly used to receive signals from each sensing module output signal, and perform simple data cleaning on the received signal information and transmit it to the computer, that is, the pre-inspection computer PC end.
  • the data acquisition system is electrically connected to the pre-in
  • the coupling adopts elastic element flexible coupling. It can compensate for the relative displacement of the two axes, compensate for the coaxiality error of the torque sensor, the motor under test and the load motor, and has the ability to buffer and absorb shock, which is especially necessary when the motor under test runs at high speed.
  • the monitoring principle of the motor status monitoring device is as follows: the four-dimensional sensing module continuously collects data signals in the electrical dimension, mechanical dimension, acoustic dimension, and temperature dimension when the motor is working. All collected data signals are received by the data acquisition board, and corresponding data preprocessing is performed on it and then transmitted to the pre-inspection computer PC. By training the deep neural network model on the PC side of the pre-inspection computer, real-time data analysis and data fusion are performed on the various dimensions of data transmitted to the PC side of the pre-inspection computer, important fault features are extracted, and corresponding diagnostic decisions are made in real time.
  • a multi-dimensional data fusion motor fault pre-inspection method includes the following pre-inspection steps
  • the multi-dimensional data fusion motor fault pre-inspection device described in Embodiment 1 collects the original data of the tested motor when it is working from four dimensions; the four dimensions are the electrical dimension, the mechanical dimension, the acoustic dimension and the temperature dimension;
  • the multi-dimensional data fusion motor fault pre-detection system described in Embodiment 2 performs data cleaning on the original data of the tested motor in four dimensions collected in the above step S1.
  • the data cleaning includes missing data and noise data.
  • a classification neural network is used to diagnose the fault of the motor under test.
  • the multi-dimensional data fusion technology is to build a neural network model of Transformer+SAE model architecture, and the multi-dimensional data after data preprocessing by the data acquisition board in the data acquisition system is input as an input sample to the training
  • multi-dimensional data fusion is performed to extract important fault feature information
  • the classification neural network uses Softmax multi-classifier as the classification network, and the important output of the Transformer+SAE model architecture is
  • the fault characteristic information is input into the trained classification neural network as an input sample.
  • the output result of the model is the probability result that the motor under test belongs to various faults. The highest probability among all output probabilities is the final motor fault diagnosis result.
  • step S2 data cleaning the data cleaning here is mainly the processing of missing values and noise data processing, which helps to improve the efficiency of deep neural network training and the performance of the final network.
  • the present invention adopts two neural networks, one performs data fusion on multi-dimensional information of the motor status to extract fault features, and the other performs fault diagnosis on the extracted fault feature information, avoiding the complex modeling process and solving the problem of limited extraction of fault feature information by a single sensor. , thus making the fault diagnosis results unreliable.
  • the present invention adopts the deep neural network model architecture of Transformer+SAE.
  • Transformer as an efficient feature extractor, has strong parallel computing capabilities and faster model training efficiency. It not only solves the problem of short-term memory of the RNN model, but also makes up for it. It overcomes the limitations of CNN's ability to capture long-distance features.
  • SAE is an unsupervised algorithm that can automatically learn features from unlabeled data and can provide better feature descriptions than the original data, thereby automatically completing the dimensionality reduction process. Through the combination of the two, it is possible to reduce the amount of calculation and improve the efficiency of data fusion, while also increasing the reliability of motor fault diagnosis results.
  • this invention uses a deep neural network to achieve reliable implementation. Through a large amount of sample data training, the neural network has good motor fault identification capabilities.
  • the overall network structure is shown in Figure 5.
  • the structure includes two parts: a multi-dimensional data fusion network and a classification neural network.
  • the dotted box on the left is The multi-dimensional data fusion network part, the dotted box on the right is the classification neural network part.
  • Pass the multi-dimensional data collected on the motor under test through their respective Transformers to extract the more important feature vectors in their respective data.
  • Multiple feature vectors are used as inputs to SAE (Stacked Autoencoder), and feature learning is continuously performed through multiple fully connected layers to achieve multi-dimensional data fusion and extract better feature descriptions.
  • SAE Stacked Autoencoder
  • feature learning is continuously performed through multiple fully connected layers to achieve multi-dimensional data fusion and extract better feature descriptions.
  • the better feature description is input into the softmax layer of the classification neural network to identify the fault features and complete fault diagnosis.
  • the multidimensional data fusion network is a neural network architecture that combines Transformer and SAE, as shown in the dotted box on the left side of Figure 5.
  • Transformer serves as an efficient feature extractor.
  • Each data signal from the PC, a total of 7, passes through a Transformer and outputs the corresponding feature vector as the input of the SAE network.
  • the Transformer model only applies the encoder layer alone, and the specific structure of the encoder is shown in Figure 6.
  • SAE also adopts a multi-layer neural network architecture, which is composed of multiple fully connected layers.
  • the latter layer is the result of the feature synthesis of the previous layer, so the number of neurons is reduced layer by layer.
  • SAE can give better results than the original data.
  • Better feature description thus automatically completing the dimensionality reduction process, that is, completing the key step of data fusion.
  • the number of neurons in the SAE output layer is the number of important features selected after data fusion.
  • the classification neural network is implemented using Softmax multi-classifier.
  • the specific network structure is shown in Figure 7.
  • the important fault characteristic information output by the Transformer+SAE model is input into the Softmax multi-classifier as an input sample.
  • Each neuron in the output layer must complete two functions: linear weighted summation and substituting the summation result into the Softmax function. calculate.
  • the m neurons on the output layer of this classification neural network essentially correspond to m health conditions of the motor, such as no motor fault, stator winding fault, air gap eccentricity fault, rolling bearing fault, rotor winding fault, rotor unbalance fault, etc. .
  • the output result of the model is essentially the probability result of the motor belonging to various health conditions. The largest probability among all output probabilities is the final motor fault diagnosis result.
  • the motor under test that is suspected to be faulty can be disassembled and installed in the multi-dimensional data fusion motor fault pre-inspection device in Embodiment 1 and the pre-inspection system in Embodiment 1 with the loaded motor assembled together. , by loading the motor and its motor drive controller to simulate the actual load condition of the motor under test to monitor the working status of the motor under test.
  • the working data information of the four-dimensional sensing detection of the original data can be quickly and accurately monitored, analyzed and judged to obtain accurate fault diagnosis results. Or in practical applications in the industrial field, there is no need to use a loaded motor, because the motor under test itself has an actual load during actual operation.
  • the timely solution of the present invention can also be used in Embodiment 1, Embodiment 2 and Embodiment 3.
  • Using the torque sensor to obtain the load size data information of the working motor can monitor and pre-check to obtain different hidden fault phenomena in motor operation, improve the accuracy and reliability of motor fault pre-check, and provide safety guarantee for timely maintenance of motor faults.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

一种多维度数据融合电机故障预检装置、系统及其预检方法，预检装置包括被测电机（10）、加载电机（20）及其各自的电机驱动控制器，以及从四个维度采集被测电机（10）工作时原始数据的四个维度传感模块，四个维度传感模块分别从电学、力学、声学和温度四个维度上进行采集被测电机（10）工作时的上述不同维度的数据信号，并将采集得到的数据信息进行数据处理和数据融合，得到故障特征后并传输至PC端（70），实现对电机（10）运行故障进行预测监测，及时监测预检获得电机（10）运行不同故障隐患现象，提高电机（10）故障预检准确性和可靠性，为电机（10）故障获得及时检修维护带来安全保证。

Description

多维度数据融合电机故障预检装置、系统及其预检方法 技术领域
本发明涉及一种电机故障预测与诊断，尤其是涉及一种使用于对电机故障预测与诊断中的多维度数据融合电机故障预检装置、系统及其预检方法。
 背景技术
电机作为重要的驱动能源，广泛应用于不同领域，在我们日常生产和生活中随处可见。如今随着电机的功能性不断完善，但跟随而来的便是其内部结构及运转技术种类也越来越复杂，由此使得设备故障几率大幅提升。而一旦电机出现故障现象，所带来的经济损失也可能将无法预测，轻则导致设备运转中断和生产力中断与损失，重则可能对人员的生命和企业资产造成威胁。而且电机出现故障后采用的传统维修策略是在有限的维护资源下，由专业人员现场做出诊断并维护，而这也很难做到电机一出现故障就能马上得到解决，随着停机时间的一再拉长，电机故障所带来的损失也就更大。可见在此情况下，对电机故障预测性的维护策略就变得尤为重要，通过对电机运行状态的监测和故障预检，提前对电机运行的健康状况做出预测判断，提前采取维修方案，保证电机可靠运行，避免带来不必要的损失。
另外，采用单一传感器所能提取故障特征信息有限，从而使得故障诊断结果并不可靠。在此情况下，利用多个不同类型传感器测得的数据进行数据融合的技术变得尤为重要，其可以提取更加精确且全面的故障特征信息，提高电机故障预检的准确率和可靠性。
 技术问题
本发明为解决现有电机在运行出现故障后，容易因无法及时得到有效对电机故障的及时检修维护而带来的停产等不同程度损失等现状而提供的一种可以对电机运行故障进行预测监测，及时监测预检获得电机运行不同故障隐患现象，提高电机故障预检准确性和可靠性，为电机故障获得及时检修维护带来安全保证的多维度数据融合电机故障预检装置、系统及其预检方法。
 技术解决方案
本发明为解决上述技术问题所采用的具体技术方案为：一种多维度数据融合电机故障预检装置，包括被测电机及其电机驱动控制器，其特征在于：还包括加载电机及其电机驱动控制器、以及从四个维度采集被测电机工作时原始数据的四个维度传感模块，四个维度传感模块分别为电学维度传感模块、力学维度传感模块、声学维度传感模块和温度维度传感模块，其中电学传感模块用于采集电机工作状态时的电压、电流和功率数据，力学传感模块用于从力学的维度采集电机工作时的振动、扭矩和转速数据，声学维度传感模块用于从声学的维度采集电机工作时发出的噪声数据，温度维度传感模块用于从温度的维度采集电机工作时的表面温度变化数据；被测电机和加载电机均设于同一工作台上，四个维度传感模块均与数据采集卡电连接。采用多维度传感模块进行采集电机工作时的多维度数据信息，可以对电机运行故障进行预测监测，及时监测预检获得电机运行不同故障隐患现象，提高电机故障预检准确性和可靠性，为电机故障获得及时检修维护带来安全保证。
作为优选，所述的力学维度传感模块采用扭矩传感器和振动传感器，其中扭矩传感器设于被测电机和加载电机相互联接的两个联轴器之间位置处；振动传感器采用磁性连接座，通过磁性连接座磁性吸附连接固定在被测电机外壳的上端处。提高对被测电机的工作时的振动监测与扭矩监测数据获得便捷可靠有效性，提高安装拆卸维护便捷性。
作为优选，所述的被测电机和加载电机均采用为出轴设计结构，力学维度传感模块中配置的扭矩传感器也采用出轴设计结构，被测电机、扭矩传感器和加载电机的出轴部分依次联接。提高满足被测电机、扭矩传感器和加载电机的工作使用同轴要求。
作为优选，所述的工作台采用T型槽工作台结构，T型槽工作台上设有被测电机安装架、加载电机安装架、扭矩传感器安装架和声学维度传感器安装架，被测电机通过被测电机安装架连接固定设于T型槽工作台上，加载电机通过加载电机安装架连接固定设于T型槽工作台上，扭矩传感器通过扭矩传感器安装架连接固定设于T型槽工作台上，声学维度传感器通过声学维度传感器安装架连接固定设于T型槽工作台上，同时保持被测电机、扭矩传感器和加载电机的出轴部分满足同轴度要求。提高电机安装架、加载电机安装架、扭矩传感器安装架和声学维度传感器安装架的安装使用与拆卸维护便捷灵活可靠性，提高安装定位便捷可靠有效性，提高同轴度要求可靠性。
作为优选，所述的温度维度传感模块采用温度传感器，温度传感器采用磁铁式贴片式温度传感器，直接吸附设置在被测电机轴承处上方的电机外壳上。提高温度维度传感模块的安装维护便捷灵活有效性。
作为优选，所述的电学维度传感模块采用采集被测电机输出的电压、电流和功率信号的功率计，功率计接收来自驱动电机控制器发出的动力信号并传输给待测电机，同时输出被测电机工作时的电压、电流和功率信号。提高被测电机的工作信号输出采集便捷可靠有效性。
本发明的另一个发明目的在于提供一种多维度数据融合电机故障预检系统，其特征在于：采用了上述技术方案之一所述的多维度数据融合电机故障预检装置，其中被测电机及其电机驱动控制器组成预检系统中的驱动系统，加载电机及其电机驱动控制器组成预检系统中的加载系统，被测电机和加载电机的输出端均配置有联轴器，两个联轴器之间配置设有扭矩传感器，两个联轴器组成传动系统，传动系统连接设于被测电机和加载电机之间，四个维度传感模块和数据采集卡组成数据采集系统，数据采集系统与预检电脑PC端电连接。采用多维度传感模块进行采集电机工作时的多维度数据信息，可以对电机运行故障进行预测监测，及时监测预检获得电机运行不同故障隐患现象，提高电机故障预检准确性和可靠性，为电机故障获得及时检修维护带来安全保证。
作为优选，所述的联轴器采用弹性元件挠性联轴器。提高对被测电机、加载电机及两个联轴器的相对位移补偿，弥补扭矩传感器、待测电机和负载电机的同轴度误差，而且具有缓冲减震的能力，在被测电机高速运转时尤为必要。
本发明的又一个发明目的在于一种多维度数据融合电机故障预检方法，其特征在于：包括如下预检步骤
S1.上述技术方案之一所述的多维度数据融合电机故障预检装置从四个维度采集被测电机工作时的原始数据；
S2. 上述技术方案之一所述的多维度数据融合电机故障预检系统对上述S1步骤所采集得到四个维度的被测电机原始数据分别进行数据清理，所述的数据清理包括对缺失数据和噪声数据的数据清理处理；
S3. 根据上述S2步骤预处理之后的多维数据，利用多维数据融合技术对被测电机的状态信息进行融合，并提取故障特征；
S4. 根据数据融合技术提取的故障特征，采用分类神经网络对被测电机进行故障诊断。
采用多维度传感模块进行采集电机工作时的多维度数据信息，可以对电机运行故障进行预测监测，及时监测预检获得电机运行不同故障隐患现象，提高电机故障预检准确性和可靠性，为电机故障获得及时检修维护带来安全保证。采用两个神经网络，一个对电机状态的多维信息进行数据融合提取故障特征，一个对提取的故障特征信息进行故障诊断，避免了复杂的建模过程，解决了单一传感器提取故障特征信息有限，从而使得故障诊断结果并不可靠问题。
作为优选，在上述S3步骤中，所述的多维数据融合技术为搭建Transformer+SAE 模型架构的神经网络模型，将经数据采集系统中的数据采集板卡进行数据预处理之后的多维数据作为输入样本输入到训练好的Transformer+SAE模型架构中，进行多维数据融合从而提取重要故障特征信息；在上述S4步骤中，所述的分类神经网络采用Softmax多分类器作为分类网络，将Transformer+SAE模型架构输出的重要故障特征信息作为输入样本输入到已训练好的分类神经网络，该模型的输出结果是该被测电机属于各种故障的概率结果，所有输出的概率中最大的概率即最终电机故障诊断结果。
采用两个神经网络，一个对电机状态的多维信息进行数据融合提取故障特征，一个对提取的故障特征信息进行故障诊断，避免了复杂的建模过程，解决了单一传感器提取故障特征信息有限，从而使得故障诊断结果并不可靠问题。
 有益效果
采用多维度传感模块进行采集电机工作时的多维度数据信息，可以对电机运行故障进行预测监测，及时监测预检获得电机运行不同故障隐患现象，提高电机故障预检准确性和可靠性，为电机故障获得及时检修维护带来安全保证。
采用两个神经网络，一个对电机状态的多维信息进行数据融合提取故障特征，一个对提取的故障特征信息进行故障诊断，避免了复杂的建模过程，解决了单一传感器提取故障特征信息有限，从而使得故障诊断结果并不可靠问题。
对于多维数据融合技术，本发明采用Transformer+SAE的深度神经网络模型架构，Transformer作为高效的特征提取器，并行计算能力强，模型训练效率更快，既解决了RNN模型短期记忆的问题，又弥补了CNN对长距离特征捕获能力的局限。而SAE一种无监督算法，可以自动从无标注数据中学习特征，可以给出比原始数据更好的特征描述，从而自动完成了降维的过程。通过两个的结合实现了既减小计算量、提高数据融合的效率的同时，也增加了电机故障诊断结果的可靠度。
 附图说明
下面结合附图和具体实施方式对本发明做进一步的详细说明。
图1是本发明多维度数据融合电机故障预检装置、系统及其预检方法的原理结构示意图。 
图2是本发明多维度数据融合电机故障预检装置的结构示意图。
图3是本发明多维度数据融合电机故障预检系统的原理结构示意图。
图4是本发明多维度数据融合电机故障预检系统的电气结构示意图。
图5本发明多维度数据融合电机故障预检方法张中多维数据融合技术的电机故障诊断网络整体结构图。
图6本发明多维度数据融合电机故障预检方法张中Transformer+SAE 模型架构示意图。
图7是本发明多维度数据融合电机故障预检方法张中Softmax多分类器网络结构示意图。
 本发明的最佳实施方式
实施例1：
图1、图2所示的实施例中，一种多维度数据融合电机故障预检装置，包括被测电机10及其电机驱动控制器，还包括加载电机20及其电机驱动控制器、以及从四个维度采集被测电机工作时原始数据的四个维度传感模块（多维数据），四个维度传感模块分别为电学维度传感模块、力学维度传感模块、声学维度传感模块和温度维度传感模块，其中电学传感模块用于从电学的维度采集电机工作状态时的电压、电流和功率数据，力学传感模块用于从力学的维度采集电机工作时的振动、扭矩和转速数据，声学维度传感模块用于从声学的维度采集电机工作时发出的噪声数据，温度维度传感模块用于从温度的维度采集电机工作时的表面温度变化数据；被测电机和加载电机均设于同一工作台80上，四个维度传感模块均与数据采集卡电连接。数据采集板卡用于接收各维度传感模块采集的信息，数据采集板卡上的处理器可处理数据，做简单的数据清理和数据融合，并传输故障特征至预检电脑PC端70。
力学维度传感模块采用扭矩传感器14和振动传感器12，其中扭矩传感器14安装连接设于被测电机10和加载电机20相互联接的两个联轴器15之间位置处；振动传感器12采用磁性连接座，通过磁性连接座磁性吸附连接固定在被测电机10外壳的上端处。振动传感器12可以同时采集到水平方向的振动信息和垂直方向的振动信息。被测电机10和加载电机20均采用为出轴设计结构，力学维度传感模块中配置的扭矩传感器14也采用出轴设计结构，被测电机10和加载电机20均的输出端均配置联轴器15，被测电机10、扭矩传感器14和加载电机20的出轴部分分别通过联轴器14实现依次联接。通过两个联轴器14可将待测电机、扭矩传感器、加载电机的出轴部分连接起来。工作台80采用T型槽工作台结构，T型槽工作台上安装设有被测电机安装架16、加载电机安装架21、扭矩传感器安装架和声学维度传感器安装架17，被测电机10通过被测电机安装架16连接固定设于T型槽工作台上，加载电机20通过加载电机安装架21连接固定设于T型槽工作台上，扭矩传感器14通过扭矩传感器安装架141连接固定设于T型槽工作台上，声学维度传感器14通过声学维度传感器安装架连接固定设于T型槽工作台上，同时保持被测电机10、扭矩传感器14和加载电机20的出轴部分满足同轴度要求。声学传感模块是指采用用于采集待测电机工作时声音信号的声音传感器13，优选地采用探头式的工业噪声传感器，通过螺母固紧于安装架上，安装于靠近待测电机20mm处的一侧。温度维度传感模块采用温度传感器11，温度传感器11采用磁铁式贴片式温度传感器，直接吸附设置在被测电机10轴承处上方的电机外壳上，方便安装与拆卸。温度维度传感模块指用于采集待测电机工作时温度信号WDXH的温度传感器11，电学维度传感模块采用采集被测电机输出的电压、电流和功率信号YLGXH的功率计，功率计接收来自驱动电机控制器发出的动力信号并传输给待测电机，同时输出被测电机工作时的电压、电流和功率信号YLGXH。
实施例
图1、图2、图3、图4所示的实施例中，一种多维度数据融合电机故障预检系统，采用了实施例1所述的多维度数据融合电机故障预检装置，其中被测电机及其电机驱动控制器组成预检系统中的驱动系统，驱动电机控制器40用于控制被测电机的输入功率，向被测电机10输入控制信号KZXH，驱动电机控制器40可以发送控制信号和动力信号给被测电机（或称为待测电机），控制被测电机按照设定的方向、速度转动；加载电机20及其电机驱动控制器组成预检系统中的加载系统，加载电机控制器50用于模拟被测电机的实际负载情况，加载电机控制器50向加载电机20输送动力信号DLXH和加载控制信号JKZXH，控制加载电机按照设定的方向、速度转动，目的是为了模拟被测电机的实际负载情况；被测电机10和加载电机20的输出端均配置有联轴器15，两个联轴器之间配置设有扭矩传感器，两个联轴器组成传动系统，传动系统连接设于被测电机和加载电机之间，四个维度传感模块和数据采集卡组成数据采集系统，数据采集系统包括电学传感模块、力学传感模块、声学传感模块、温度传感模块和数据采集板卡60，电学传感模块包括用于采集待测电机输出的电压、电流、功率信号YLGXH的功率计30，功率计可接收来自驱动电机控制器发出的动力信号并传输给待测电机，同时输出电机工作时的电压、电流、功率信号；力学传感模块包括用于采集待测电机振动信号ZDXH的振动传感器和用于采集待测电机扭矩、转速信号NJXH的扭矩传感器；声学传感模块包括用于采集待测电机工作时声音信号SYXH的声音传感器；温度传感模块包括用于采集待测电机工作时温度信号的温度传感器；数据采集板卡主要用于接收来自各个传感模块的输出信号，并对接收到的信号信息分别进行简单的数据清理并传输至计算机即预检电脑PC端，数据采集系统与预检电脑PC端70电连接。
联轴器采用弹性元件挠性联轴器。可以补偿两轴的相对位移，弥补扭矩传感器、待测电机和负载电机的同轴度误差，而且具有缓冲减震的能力，在待测电机高速运转时尤为必要。
电机状态监测装置的监测原理如下：通过四个维度传感模块连续采集电机工作时电学维度、力学维度、声学维度、温度维度这四个维度的数据信号。所有采集得到的数据信号由数据采集板卡接收，在其上做相应的数据预处理之后并传输至预检电脑PC端。通过预检电脑PC端上训练好深度神经网络模型，对传输至预检电脑PC端的各个维度数据进行实时的数据分析、数据融合、提取重要故障特征、实时做出相应的诊断决策。
实施例
图1、图2、图3、图4、图5、图6、图7所示的实施例中，一种多维度数据融合电机故障预检方法，包括如下预检步骤
      S1.实施例1所述的多维度数据融合电机故障预检装置从四个维度采集被测电机工作时的原始数据；四个维度分别为电学维度、力学维度、声学维度和温度维度；
S2.实施例2所述的多维度数据融合电机故障预检系统对上述S1步骤所采集得到四个维度的被测电机原始数据分别进行数据清理，所述的数据清理包括对缺失数据和噪声数据的数据清理处理；
S3. 根据上述S2步骤预处理之后的多维数据，利用多维数据融合技术对被测电机的状态信息进行融合，并提取故障特征；
S4. 根据数据融合技术提取的故障特征，采用分类神经网络对被测电机进行故障诊断。
在上述S3步骤中，所述的多维数据融合技术为搭建Transformer+SAE 模型架构的神经网络模型，将经数据采集系统中的数据采集板卡进行数据预处理之后的多维数据作为输入样本输入到训练好的Transformer+SAE模型架构中，进行多维数据融合从而提取重要故障特征信息；在上述S4步骤中，所述的分类神经网络采用Softmax多分类器作为分类网络，将Transformer+SAE模型架构输出的重要故障特征信息作为输入样本输入到已训练好的分类神经网络，该模型的输出结果是该被测电机属于各种故障的概率结果，所有输出的概率中最大的概率即最终电机故障诊断结果。
进一步的，上述步骤S2数据清理，此处的数据清理主要是缺失值的处理和噪声数据处理，有助于提高深度神经网络训练的效率以及最终网络的性能。
本发明采用两个神经网络，一个对电机状态的多维信息进行数据融合提取故障特征，一个对提取的故障特征信息进行故障诊断，避免了复杂的建模过程，解决了单一传感器提取故障特征信息有限，从而使得故障诊断结果并不可靠问题。
对于多维数据融合技术，本发明采用Transformer+SAE的深度神经网络模型架构，Transformer作为高效的特征提取器，并行计算能力强，模型训练效率更快，既解决了RNN模型短期记忆的问题，又弥补了CNN对长距离特征捕获能力的局限。而SAE一种无监督算法，可以自动从无标注数据中学习特征，可以给出比原始数据更好的特征描述，从而自动完成了降维的过程。通过两个的结合实现了既减小计算量、提高数据融合的效率的同时，也增加了电机故障诊断结果的可靠度。
为实现基于多维数据融合电机故障的诊断，本次发明采用搭建深度神经网络来可靠实现。通过大量的样本数据训练，让该神经网络具备良好的电机故障判别的能力，其网络整体结构如图5所示, 该结构包括多维数据融合网络和分类神经网络两个部分，左边的虚线框是多维数据融合网络部分，右边的虚线框是分类神经网络部分。将待测电机上采集到的多维数据，分别经过各自的Transformer，提取出各自数据中较重要的特征向量。将多个特征向量作为SAE（堆叠自编码器）的输入，经过多个全连接层，不断的进行特征学习，实现多维数据融合，提取出更好的特征描述。最后将更好的特征描述输入分类神经网络softmax层，对故障特征进行判别，完成故障诊断。
多维数据融合网络是Transformer与SAE结合的神经网络架构，如图5左边的虚线框所示。Transformer作为高效的特征提取器，每一个来自PC端的数据信号，总共7个，都经过一个Transformer，输出相应的特征向量，使其作为SAE网络的输入。本发明中Transformer模型只单独应用encoder层，encoder具体结构如图6所示。在原本的encoder结构中做了些变化，从原来一个Multi-Head Attention层、 Add & Norm层、Feed Forward层，分别增加到两个。此举的目的旨在加深神经网络，使最后的电机故障特征信息的提取更加全面。同时，SAE也采用多层的神经网络架构，其由多个全连接层组合，后一层是前一层特征综合的结果，因此逐层神经元数量减少，也是因为SAE可以给出比原始数据更好的特征描述，从而自动完成了降维的过程，即完成数据融合的关键步骤，SAE输出层的神经元个数即数据融合之后选出的重要特征的个数。
分类神经网络采用Softmax多分类器来实现，具体的网络结构如图7所示。将Transformer+SAE模型输出的重要故障特征信息作为输入样本输入Softmax多分类器中,该输出层的每个神经元都要完成两个功能分别是线性加权求和以及对求和结果代入Softmax函数进行计算。该分类神经网络输出层上的m个神经元，实质上分别对应电机的m种健康状况，例如电机无故障、定子绕组故障、气隙偏心故障、滚动轴承故障、转子绕组故障、转子不平衡故障等。通过softmax函数的处理，模型的输出结果实质上是电机属于各种健康状况的概率结果，所有输出的概率中最大的概率即最终电机故障诊断的结果。
使用上，可以是将怀疑有故障问题的被测电机拆卸下来后安装使用在有加载电机组装在一起的实施例1中的多维度数据融合电机故障预检装置及实施例1的预检系统中，进行通过加载电机及其电机驱动控制器来模拟被测电机的实际负载情况进行监测被测电机工作状态时原始数据四个维度传感检测的工作数据信息，进而可以快速准确监测分析判断得到准确的故障诊断结果。或者是在工业领域实际应用中，则无使用需加载电机，因为被测电机在实际运行过程中本身就有实际负载，因而使用本发明及时方案实施例1、实施例2和实施例3同样可以用扭矩传感器获取工作电机的负载大小数据信息，可以监测预检获得电机运行不同故障隐患现象，提高电机故障预检准确性和可靠性，为电机故障获得及时检修维护带来安全保证。

Claims (10)

  1. 一种多维度数据融合电机故障预检装置，包括被测电机及其电机驱动控制器，其特征在于：还包括加载电机及其电机驱动控制器、以及从四个维度采集被测电机工作时原始数据的四个维度传感模块，四个维度传感模块分别为电学维度传感模块、力学维度传感模块、声学维度传感模块和温度维度传感模块，其中电学传感模块用于采集电机工作状态时的电压、电流和功率数据，力学传感模块用于从力学的维度采集电机工作时的振动、扭矩和转速数据，声学维度传感模块用于从声学的维度采集电机工作时发出的噪声数据，温度维度传感模块用于从温度的维度采集电机工作时的表面温度变化数据；被测电机和加载电机均设于同一工作台上，四个维度传感模块均与数据采集卡电连接。
  2. 按照权利要求1所述的多维度数据融合电机故障预检装置，其特征在于：所述的力学维度传感模块采用扭矩传感器和振动传感器，其中扭矩传感器设于被测电机和加载电机相互联接的两个联轴器之间位置处；振动传感器采用磁性连接座，通过磁性连接座磁性吸附连接固定在被测电机外壳的上端处。
  3. 按照权利要求1所述的多维度数据融合电机故障预检装置，其特征在于：所述的被测电机和加载电机均采用为出轴设计结构，力学维度传感模块中配置的扭矩传感器也采用出轴设计结构，被测电机、扭矩传感器和加载电机的出轴部分依次联接。
  4. 按照权利要求1所述的多维度数据融合电机故障预检装置，其特征在于：所述的工作台采用T型槽工作台结构，T型槽工作台上设有被测电机安装架、加载电机安装架、扭矩传感器安装架和声学维度传感器安装架，被测电机通过被测电机安装架连接固定设于T型槽工作台上，加载电机通过加载电机安装架连接固定设于T型槽工作台上，扭矩传感器通过扭矩传感器安装架连接固定设于T型槽工作台上，声学维度传感器通过声学维度传感器安装架连接固定设于T型槽工作台上，同时保持被测电机、扭矩传感器和加载电机的出轴部分满足同轴度要求。
  5. 按照权利要求1所述的多维度数据融合电机故障预检装置，其特征在于：所述的温度维度传感模块采用温度传感器，温度传感器采用磁铁式贴片式温度传感器，直接吸附设置在被测电机轴承处上方的电机外壳上。
  6. 按照权利要求1所述的多维度数据融合电机故障预检装置，其特征在于：所述的电学维度传感模块采用采集被测电机输出的电压、电流和功率信号的功率计，功率计接收来自驱动电机控制器发出的动力信号并传输给待测电机，同时输出被测电机工作时的电压、电流和功率信号。
  7. 一种多维度数据融合电机故障预检系统，其特征在于：采用了权利要求1～6之一所述的多维度数据融合电机故障预检装置，其中被测电机及其电机驱动控制器组成预检系统中的驱动系统，加载电机及其电机驱动控制器组成预检系统中的加载系统，被测电机和加载电机的输出端均配置有联轴器，两个联轴器之间配置设有扭矩传感器，两个联轴器组成传动系统，传动系统连接设于被测电机和加载电机之间，四个维度传感模块和数据采集卡组成数据采集系统，数据采集系统与预检电脑PC端电连接。
  8. 按照权利要求1所述的多维度数据融合电机故障预检系统，其特征在于：所述的联轴器采用弹性元件挠性联轴器。
  9. 一种多维度数据融合电机故障预检方法，其特征在于：包括如下预检步骤
    S1.权利要求1～6之一所述的多维度数据融合电机故障预检装置从四个维度采集被测电机工作时的原始数据；
    S2.权利要求7～8之一所述的多维度数据融合电机故障预检系统对上述S1步骤所采集得到四个维度的被测电机原始数据分别进行数据清理，所述的数据清理包括对缺失数据和噪声数据的数据清理处理；
    S3. 根据上述S2步骤预处理之后的多维数据，利用多维数据融合技术对被测电机的状态信息进行融合，并提取故障特征；
    S4. 根据数据融合技术提取的故障特征，采用分类神经网络对被测电机进行故障诊断。
  10. 按照权利要求9所述的多维度数据融合电机故障预检系统，其特征在于：在上述S3步骤中，所述的多维数据融合技术为搭建Transformer+SAE 模型架构的神经网络模型，将经数据采集系统中的数据采集板卡进行数据预处理之后的多维数据作为输入样本输入到训练好的Transformer+SAE模型架构中，进行多维数据融合从而提取重要故障特征信息；在上述S4步骤中，所述的分类神经网络采用Softmax多分类器作为分类网络，将Transformer+SAE模型架构输出的重要故障特征信息作为输入样本输入到已训练好的分类神经网络，该模型的输出结果是该被测电机属于各种故障的概率结果，所有输出的概率中最大的概率即最终电机故障诊断结果。
PCT/CN2023/103024 2022-07-11 2023-06-28 多维度数据融合电机故障预检装置、系统及其预检方法 WO2024012199A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210811411.3A CN115308593A (zh) 2022-07-11 2022-07-11 多维度数据融合电机故障预检装置、系统及其预检方法
CN202210811411.3 2022-07-11

Publications (1)

Publication Number Publication Date
WO2024012199A1 true WO2024012199A1 (zh) 2024-01-18

Family

ID=83857449

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/103024 WO2024012199A1 (zh) 2022-07-11 2023-06-28 多维度数据融合电机故障预检装置、系统及其预检方法

Country Status (2)

Country Link
CN (1) CN115308593A (zh)
WO (1) WO2024012199A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117871096A (zh) * 2024-03-11 2024-04-12 昆明理工大学 一种滚动轴承故障模拟实验装置及故障在线诊断方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115308593A (zh) * 2022-07-11 2022-11-08 浙江联宜电机有限公司 多维度数据融合电机故障预检装置、系统及其预检方法
CN115683235B (zh) * 2022-12-28 2023-03-28 常州检验检测标准认证研究院 一种工业机器人伺服电机振动故障检测方法及装置
CN116910627B (zh) * 2023-09-11 2023-11-17 四川大学 一种提高电动汽车电驱系统效率的方法、系统及存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1611955A (zh) * 2003-05-17 2005-05-04 杜玉晓 分布式智能电机监测系统
CN201188125Y (zh) * 2008-03-17 2009-01-28 上海海事大学 一种多数据采集方式支持的电机特性测试系统
CN108921303A (zh) * 2018-05-29 2018-11-30 青岛鹏海软件有限公司 工业电动机的故障诊断及预测性维护方法
CN109145886A (zh) * 2018-10-12 2019-01-04 西安交通大学 一种多源信息融合的异步电机故障诊断方法
KR20200091073A (ko) * 2019-01-22 2020-07-30 주식회사 케이디파워 발전기의 고장예측 진단 시스템
CN113533962A (zh) * 2021-07-29 2021-10-22 上海交通大学 基于多物理信号传感器决策融合的感应电机健康诊断系统
CN115308593A (zh) * 2022-07-11 2022-11-08 浙江联宜电机有限公司 多维度数据融合电机故障预检装置、系统及其预检方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1611955A (zh) * 2003-05-17 2005-05-04 杜玉晓 分布式智能电机监测系统
CN201188125Y (zh) * 2008-03-17 2009-01-28 上海海事大学 一种多数据采集方式支持的电机特性测试系统
CN108921303A (zh) * 2018-05-29 2018-11-30 青岛鹏海软件有限公司 工业电动机的故障诊断及预测性维护方法
CN109145886A (zh) * 2018-10-12 2019-01-04 西安交通大学 一种多源信息融合的异步电机故障诊断方法
KR20200091073A (ko) * 2019-01-22 2020-07-30 주식회사 케이디파워 발전기의 고장예측 진단 시스템
CN113533962A (zh) * 2021-07-29 2021-10-22 上海交通大学 基于多物理信号传感器决策融合的感应电机健康诊断系统
CN115308593A (zh) * 2022-07-11 2022-11-08 浙江联宜电机有限公司 多维度数据融合电机故障预检装置、系统及其预检方法

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117871096A (zh) * 2024-03-11 2024-04-12 昆明理工大学 一种滚动轴承故障模拟实验装置及故障在线诊断方法

Also Published As

Publication number Publication date
CN115308593A (zh) 2022-11-08

Similar Documents

Publication Publication Date Title
WO2024012199A1 (zh) 多维度数据融合电机故障预检装置、系统及其预检方法
Chao et al. Adaptive decision-level fusion strategy for the fault diagnosis of axial piston pumps using multiple channels of vibration signals
CN107220469B (zh) 一种风机状态估计的方法及系统
WO2019063692A1 (en) SYSTEM, METHOD AND CONTROL UNIT FOR THE DIAGNOSIS AND PREDICTION OF THE LIFETIME OF ONE OR MORE ELECTROMECHANICAL SYSTEMS
KR20200014129A (ko) 딥러닝을 이용한 변압기 진단 방법
CN112285554B (zh) 基于信息融合的永磁同步电机退磁故障诊断方法与装置
Iqbal et al. CNC machine-bearing fault detection based on convolutional neural network using vibration and acoustic signal
CN115683235B (zh) 一种工业机器人伺服电机振动故障检测方法及装置
PETERKA et al. DIAGNOSTICS OF AUTOMATED TECHNOLOGICAL DEVICES.
Wang et al. Fault detection for motor drive control system of industrial robots using CNN-LSTM-based observers
Immovilli et al. Automated bearing fault detection via long short-term memory networks
CN111795819A (zh) 一种融合振动与电流信号协同学习的齿轮箱故障诊断方法
Liu et al. Rolling bearing fault diagnosis of train running gear based on optimized deep residual network
Wu et al. A review of fault diagnosis methods of robot joint servo system
CN115144751B (zh) 一种牵引电机故障诊断装置及方法
Pengbo et al. Recent progress and prospective evaluation of fault diagnosis strategies for electrified drive powertrains: A comprehensive review
Demetgul et al. Misalignment detection on linear feed axis using sensorless motor current signals
El Mahdi et al. Real Time Assessment of Novel Predictive Maintenance System Based on Artificial Intelligence for Rotating Machines.
Yeung et al. A general-purpose anomalous scenario synthesizer for rotary equipment
Wu et al. Discovering new prognostic features for the harmonic reducer in remaining useful life prediction
Shuai Research on Fault Diagnosis Technology of Industrial Robot Operation Based on Deep Belief Network
Tang et al. Fault mode recognition of planetary gears based on CNN and transfer learning
Xiaodong et al. Fault diagnosis method for multiple fault system based on multi-source information fusion
Waters et al. Vibration Anomaly Detection using Deep Autoencoders for Smart Factory
Immovilli et al. Experimental evidence of MCSA for the diagnosis of ball-bearings

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: 23838693

Country of ref document: EP

Kind code of ref document: A1