A machine learning based motor fault detection system and method

Abstract

The application discloses a kind of motor fault detection system and method based on machine learning, comprising: host computer, at least one time sequence database is configured, for reading the data of multiple parameters (including vibration parameter, temperature parameter) of storage acquisition device;Acquisition device, for collecting multiple parameters of multiple motors on motor detection pipeline, including motor barcode input equipment, PLC, IOLink conversion Profinet module and vibration temperature sensor, the sensor is adsorbed on the surface of motor to be detected by magnetic support;Processor contains computer program, reads data from at least one time sequence database to execute fault detection;Host computer is also equipped with the configuration database of storage acquisition device and motor type communication information;Host computer reads motor barcode and obtains model information, and motor model information and multiple parameter information are bound and stored in time sequence database.The application has self-adaptability, can make early warning to motor fault, can improve the production efficiency of workshop, quickly adapt complex motor production line.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of motor fault detection, and particularly relates to a motor fault detection system and method based on machine learning. BACKGROUND

[0002] At present, the traditional motor fault detection is through the following ways: 1, vibration analysis: through the vibration sensor or vibration meter to analyze the vibration of the motor, to detect the frequency, amplitude and form of vibration. Different fault modes will produce specific vibration characteristics, such as bearing failure, imbalance, looseness, etc., which can be detected and diagnosed by vibration analysis; 2, heat detection: using infrared thermometer or temperature sensor to detect the temperature of each part of the motor. Abnormal temperature rise may indicate that the motor is overloaded, winding problem, cooling system failure, etc.; 3, insulation test: measure the insulation resistance of the motor by using the insulation tester to evaluate the insulation condition of the motor. The decrease of insulation resistance may indicate that the winding has insulation damage or moisture problem; 4, phase current analysis: by measuring the current of each phase of the motor, analyzing the change of current waveform and amplitude. Abnormal current characteristics may indicate that the motor has interphase short circuit, broken wire or imbalance problem; 5, sound analysis: using sound sensor or microphone to detect the sound signal during the operation of the motor, abnormal sound characteristics may indicate that the motor has bearing wear, gear failure, etc. problem; 6, visual inspection: disassemble the motor, visually inspect the internal components such as bearings, windings, brushes, etc., to observe whether there is wear, corrosion, fracture or burnout problem.

[0003] For the above motor fault detection, the main principle is to predict from the following aspects: 1, data collection: collect data related to the state of the motor, such as current, voltage, vibration, temperature, etc. Ensure that the data set contains data samples under normal and abnormal working conditions; 2, feature extraction: extract features from the collected data; 3, data labeling: according to the actual state of the motor, label the data. Classify and mark the data samples of normal and abnormal working conditions; 4, detect the state of the motor: predict the state of the new motor data to determine whether the motor is normal or abnormal.

[0004] The above feature extraction and data labeling are more dependent on manual operation, and the features and data are labeled only when the motor has or is about to have a fault. The process has a high degree of manual intervention, is time-consuming and laborious, and the effect is poor.

[0005] Corresponding to the above motor fault detection, for example, patent No. CN202141923U, the name is "motor assembly line data acquisition and processing system" is disclosed, "monitoring center is equipped with data comparison module, according to the data comparison result, the automatic alarm function is started. When detecting the product number, the number of scrap, the number of parts, etc. Don't match, the system automatically alarms." The detection involved in this technology is essentially data comparison, and the standard value of the alarm is determined manually. When the actual value collected is not equal to the standard value, the alarm is given. It does not involve feature extraction and data labeling using machine learning method for the collected data.

[0006] In machine learning, feature extraction refers to extracting meaningful features or characteristics from raw data to better describe and represent the properties of data. Data labeling refers to providing training data with label or category information for machine learning algorithms. The process of data labeling is to tell the algorithm the correct output result or category corresponding to each input data, so that the algorithm can learn the mapping relationship between input and output. SUMMARY

[0007] Therefore, in order to solve at least one of the above defects or improvement needs of the prior art, the feature extraction and data labeling are carried out when the motor has or is about to have a fault, which is time-consuming and laborious and has poor effect. The present application provides a motor fault detection system and method based on machine learning.

[0008] The present application provides a motor fault detection system based on machine learning, characterized in that it comprises:

[0009] The host computer is configured to have at least one time series database for reading data stored from the acquisition device, the plurality of parameters including vibration parameters, temperature parameters; the acquisition device is used to acquire the plurality of parameters of the plurality of motors arranged on the motor detection pipeline;

[0010] The processor is configured to have a computer program for reading the data from the at least one time series database and executing fault detection processing;

[0011] The host computer is also configured to have a configuration database for storing communication information between the acquisition device and the motor model;

[0012] When the host computer reads the current motor barcode, the host computer acquires the motor model information from the motor barcode, and binds and stores the motor model information and the plurality of parameter information in the at least one time series database.

[0013] Further, the acquisition device comprises:

[0014] A motor barcode input device is arranged on each motor detection assembly line, and is used for scanning a motor barcode arranged in a bottom plate of a motor support, so as to read the current motor barcode;

[0015] A PLC is used for interacting with the sensor, and collecting the plurality of parameters of the motor to be detected;

[0016] A sensor is arranged on the motor to be detected, and is used for collecting the plurality of parameters and transmitting to the PLC;

[0017] The motor barcode input device sends the input motor barcode to an upper computer in real time.

[0018] Further,

[0019] The configuration database stores the motor type and the communication information corresponding to the PLC;

[0020] The PLC and the sensor further comprise a protocol conversion module.

[0021] Further, the sensor is connected to the surface of the motor to be detected by a magnetic support.

[0022] Further, the sensor is a vibration temperature sensor.

[0023] The application further provides a motor fault detection method based on machine learning, based on the above motor fault detection system, and the method comprises the following steps:

[0024] Collecting the plurality of parameters of a plurality of motors arranged on a motor detection assembly line;

[0025] Reading a current motor barcode, obtaining motor type information from the motor barcode, and binding and storing the motor type information and the plurality of parameter information in a time sequence database;

[0026] Reading, in a first file format, the collection time, vibration parameter information and temperature parameter information corresponding to the current motor type stored in the time sequence database, loading into the computer program, and repeatedly loading and analyzing to obtain fault data.

[0027] Further, the loading and analyzing to obtain fault data comprises the following steps:

[0028] Defining a parameter data input class and a parameter data prediction data class, and specifying that the time sequence database is loaded in the first file format, and the data column related to fault detection includes a collection time column, a vibration parameter column and a temperature parameter column;

[0029] initializing an MLContext object to create an ML.NET environment shared between model creation workflow objects;

[0030] transforming the data column into an IDataView object to execute the SR-CNN algorithm to analyze abnormal data and mark.

[0031] Further, the first file format is CSV format.

[0032] The application further discloses a motor fault detection method based on machine learning, characterized by comprising the following steps:

[0033] S0, configuring motor model and PLC communication information in a configuration database;

[0034] S1, acquiring motor bar code input by a motor bar code input device through a host computer;

[0035] S2, obtaining motor model information through the motor bar code;

[0036] S3, reading vibration and temperature data of the motor from the PLC;

[0037] S4, storing collection time, motor model information, vibration data and temperature data into a time series database;

[0038] S5, reading out the collection time and vibration data of the current motor model from the time series database and generating a CSV file;

[0039] S6, loading the CSV file into motor fault detection software;

[0040] S7, defining vibration data input class and vibration data prediction data class, and specifying the time column and vibration data column in the CSV file to be loaded by using the LoadColumn attribute;

[0041] S8, initializing an MLContext object to create an ML.NET environment shared between model creation workflow objects;

[0042] S9, loading data obtained in the step S7 into memory and transforming the data into an IDataView object;

[0043] S10, determining the periodic value of the time series by using the DetectSeasonality function;

[0044] S11, finding vibration abnormal data and marking by using the DetectEntireAnomalyBySrCnn method;

[0045] S12, read out and generate a CSV file from the time sequence database first the current motor model acquisition time, temperature data, repeat the above steps S6~S11 find out the temperature abnormal data and mark.

[0046] Overall, compared with the prior art, the above technical solutions of the present application can achieve the following beneficial effects:

[0047] By combining hardware and software, the machine learning is used to detect motor faults, which can adapt to the mixed production of different types of motors in the motor workshop of the vehicle manufacturer, realize 7*24 hour monitoring of the motor, continuously correct the fault detection standard combined with historical data, realize the fault alarm of the motor no longer relying on the experience of maintenance personnel, compared with the traditional motor fault alarm scheme, the present application has self-adaptability, can make early warning for motor faults, can improve the production efficiency of the workshop, and can quickly adapt to the increasingly complex motor production line.

[0048] In the detection end of the motor data, the detection data of the motors of the multiple flow lines collected by the upper computer in the application are connected with the framework of machine learning, so that multiple motors can automatically and orderly carry out fault data detection, can complete the detection of motor fault data in large quantities, and the definition of the class of data training can be performed at the model end, which can improve the expandability of motor data training. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 is a system composition schematic diagram for detecting motor faults by machine learning of the application;

[0050] Figure 2 is a system working method flow chart for detecting motor faults by machine learning of the application;

[0051] Figure 3 is a model information table of the RFID system for detecting motor faults by machine learning of the application;

[0052] Figure 4 is a PLC point address table for detecting motor faults by machine learning of the application.

[0053] 11-motor barcode input device, 12-upper computer, 13-time sequence database, 14-motor fault detection software, 15-Siemens PLC, 16-IOLink to Profinet module, 17-vibration temperature sensor, 18-configuration database. DETAILED DESCRIPTION

[0054] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0055] To achieve the above object, according to one aspect of the present application, as shown in the accompanying drawings and embodiments, the present application provides a system for detecting motor faults by machine learning, which comprises a motor barcode input device 11, a host computer 12, a time series database 13, a motor fault detection software 14, a Siemens PLC 15, an IO Link to Profinet module 16, a vibration temperature sensor 17, and a configuration database 18. Figure 1

[0056] One embodiment of the present application discloses a motor fault detection system based on machine learning, which comprises:

[0057] A motor operating parameter acquisition module is configured to acquire a plurality of parameter information of the motor, including vibration parameter information and temperature parameter information.

[0058] A data storage and detection module is configured to analyze the vibration parameter information and the temperature parameter information to obtain motor fault data.

[0059] The motor operating parameter acquisition module comprises a motor barcode input device 11, a Siemens PLC 15, an IO Link to Profinet module 16, and a vibration temperature sensor 17 connected in sequence.

[0060] The data storage and detection module comprises a time series database 13, a motor fault detection software 14, and a configuration database 18 arranged on the host computer 12.

[0061] The motor barcode input device 11 sends the input motor barcode to the host computer 12 in real time.

[0062] The configuration database 18 stores motor models and Siemens PLC communication information.

[0063] The vibration temperature sensor 17 is connected to the surface of the motor by a magnetic support.

[0064] The IO Link to Profinet module 16 converts the IO Link protocol of the vibration temperature sensor 17 into the Profinet protocol.

[0065] Further, the Siemens PLC 15 collects vibration and temperature data in the IO Link to Profinet module 16.

[0066] ​Further, the time series database 13 stores motor model, vibration, temperature data.

[0067] The motor fault detection software 14 performs fault detection according to the time series data through the SR-CNN algorithm of machine learning.

[0068] The embodiment in the application also discloses a motor fault detection method based on machine learning, comprising the following steps:

[0069] S0, configuring motor model and information of Siemens PLC communication in the configuration database 18;

[0070] S1, acquiring motor bar code input by the motor bar code input device 11 through the host computer 12;

[0071] S2, obtaining motor model information through the motor bar code;

[0072] S3, reading vibration and temperature data of the motor from the Siemens PLC 15;

[0073] S4, storing the collection time, motor model information, vibration data and temperature data to the time series database 13;

[0074] S5, reading out the collection time and vibration data of the current motor model from the time series database 13 and generating a CSV file;

[0075] S6, loading the CSV file to the motor fault detection software 14;

[0076] S7, defining vibration data input class and vibration data prediction data class, and using the LoadColumn attribute to specify the time column and vibration data column in the CSV data set that should be loaded;

[0077] S8, initializing the MLContext object, creating a new ML.NET environment, and sharing the environment between model creation workflow objects;

[0078] S9, loading vibration time series data into memory and converting it into an IDataView object;

[0079] S10, using the DetectSeasonality function to determine the period value of the time series;

[0080] S11, using the DetectEntireAnomalyBySrCnn method to find vibration abnormal data and mark it as motor vibration abnormality;

[0081] S12, generating a CSV file of temperature time series data, repeating the above method to find temperature abnormal data, and marking it as motor temperature abnormality.

[0082] The motor bar code input device 11 inputs the motor bar code to the host computer 12. In a specific embodiment, each motor in the automobile factory has a unique motor number and a derived number. The motor number has 12 bits, and the derived number has 9 bits. The feature code composed of the first to fifth bits of the motor number and the second to seventh bits of the derived number can uniquely determine the motor type.

[0083] The motor number, the derived number, the motor feature code, the derived code feature code, and the motor type information exist in the RFID system in the motor workshop. When the motor is put on line, the RFID system writes the motor number and the derived number into the chip at the bottom of the motor tray through the RFID read-write head at the first work station. When the tray flows to the work station along the production line, the RFID read-write head reads out the motor number and the derived number in the tray chip, analyzes the motor feature code and the derived number feature code through a predetermined rule, and then queries the motor type information in the database of the RFID system.

[0084] The motor type information table [t_EngineType] of the RFID system stores the motor type information, including the motor type category, the motor type name, the motor feature code, the motor number feature code, and the derived number feature code, as shown in Figure 3 .

[0085] The PLC point address table [T_PLCItems] stores the addresses of the vibration and temperature data read from the Siemens PLC, as shown in Figure 4 .

[0086] The timing database 13 and the motor fault detection software 14 are installed in the host computer 12. The timing database 13 is used to store the motor type, vibration, and temperature data. The motor fault detection software 14 uniquely determines the motor type according to the motor bar code, binds and stores the vibration and temperature data of the motor to the timing database 13,

[0087] The motor fault detection software 14 reads the data from the timing database 13, and then performs fault detection according to the timing data.

[0088] The computer program set in the processor, the motor fault detection software 14 can be integrated and set in the host computer 12.

[0089] The Siemens PLC 15 is used to collect the vibration and temperature data in the IOLink to Profinet module 16, and transmit them to the motor fault detection software 14. The IOLink to Profinet module 16 is used to convert the IOLink protocol of the vibration and temperature sensor 17 to the Profinet protocol. The vibration and temperature sensor 17 is adsorbed on the surface of the motor through a customized magnetic bracket to obtain its vibration and temperature data. The configuration database 18 is used to store the motor type configuration.

[0090] As shown in Figure 2As shown, the application further discloses a working method of a device for detecting motor failure by machine learning, comprising the following steps:

[0091] S0: configure the motor type and Siemens PLC communication information in the configuration database, and the host computer software obtains the motor number and derived number in the motor tray chip through the RFID read-write head.

[0092] S1: Obtain motor barcode

[0093] S2: Obtain motor type information through motor barcode, and obtain motor type information by comparing motor type information stored in the model information table [t_EngineType] of the RFID system, which will not be repeated.

[0094] S3: Read the vibration and temperature data of the motor from the Siemens PLC.

[0095] S4: Store the collection time, motor type information, vibration data and temperature data into the time series database.

[0096] S5: Read out the collection time and vibration data of the current motor type from the time series database and generate a CSV file; the vibration and temperature sensor is adsorbed on the surface of the motor through a magnetic support, and is connected to the Profinet network module through an IOLink communication line. The vibration and temperature sensor automatically collects the vibration and temperature data of the motor, and transmits them to the Profinet network module through the IOLink protocol. The Profinet network module is a relay station, which converts the IOLink protocol data into Ethernet protocol data. The Siemens PLC is connected to the Profinet network module through Ethernet, so that it can receive the vibration and temperature data in Ethernet protocol. The host computer reads the vibration and temperature data from the specified address in the PLC. Because the motor barcode has been obtained through the RFID read-write head at the work station, the vibration and temperature data of the motor can be stored in the time series database after being bound.

[0097] S6: Load the CSV file into the host computer software.

[0098] The time series database uses InfluxDB, which is an open source distributed time series database specially used for storing and processing time series data, and the specific form is as follows:

[0099] Time, motor barcode, vibration data, temperature data.

[0100] S7: Define the vibration data input class and vibration data prediction data class, and specify the time column and vibration data column in the CSV data set to be loaded by using the LoadColumn attribute.

[0101] S8: Initialize the MLContext object, create a new ML.NET environment, share the environment between model creation workflow objects.

[0102] S9: Load the vibration time series data into memory and convert it to an IDataView object.

[0103] S10: Use the DetectSeasonality function to determine the periodic value of the time series.

[0104] S11: Use the DetectEntireAnomalyBySrCnn algorithm to find vibration anomaly data and label it as motor vibration anomaly.

[0105] S12: Then generate a CSV file of the temperature time series data, repeat the above method to find temperature anomaly data, and label it as motor temperature anomaly.

[0106] The content described in the specification is only an example of the present application, and those skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, as long as they do not deviate from the content of the specification or exceed the scope defined by the claims.

Claims

1. A machine learning-based motor fault detection system, characterized in that, include: The host computer is configured to have at least one time-series database for reading data that stores multiple parameters from the acquisition device, including vibration parameters and temperature parameters. The acquisition device is used to acquire the multiple parameters of multiple motors installed on the motor testing production line; The data acquisition device includes: a motor barcode input device, a PLC, a protocol conversion module, and a vibration and temperature sensor; the motor barcode input device is used to scan the motor barcode to obtain the current motor model information; the vibration and temperature sensor is attached to the surface of the motor to be tested by a magnetic bracket, collects vibration and temperature parameters, and transmits them to the PLC via the protocol conversion module; The processor is configured to have a computer program for reading the data from the at least one time-series database, performing fault detection processing, specifically initializing an MLContext object through the ML.NET environment, converting the data into an IDataView object, analyzing abnormal data using the SR-CNN algorithm, generating a motor vibration fault warning based on the vibration abnormal data, and generating a motor temperature abnormal warning based on the temperature abnormal data. The host computer is also configured to have a configuration database for storing communication information between the acquisition device and the motor model; When the host computer reads the current motor barcode, it obtains the motor model information from the motor barcode and stores the motor model information and the multiple parameters in the at least one time-series database, binding them with the collection time.

2. The machine learning-based motor fault detection system according to claim 1, characterized in that, The motor barcode input device is installed on each of the motor testing production lines and is used to scan the motor barcode installed in the chassis supporting the motor to read the current motor barcode. The PLC is used to interact with the vibration temperature sensor and to collect the multiple parameters of the motor to be tested through the protocol conversion module. The motor barcode input device sends the input motor barcode to the host computer in real time.

3. A machine learning-based motor fault detection method, based on the motor fault detection system as described in any one of claims 1-2, characterized in that, Includes the following steps: Collect the aforementioned parameters of multiple motors installed on the motor testing production line; Read the current motor barcode, obtain the motor model information from the motor barcode, and bind and store the motor model information and the multiple parameters in the time-series database; The system reads the acquisition time, vibration parameter information, and temperature parameter information corresponding to the current motor model from the time-series database in the first file format, loads them into the computer program, and parses them multiple times to obtain fault data. The fault data includes abnormal vibration data and abnormal temperature data, and generates corresponding motor vibration fault warnings and motor temperature fault warnings.

4. The machine learning-based motor fault detection method as described in claim 3, characterized in that, The process of loading, parsing, and obtaining fault data includes the following steps: Define a parameter data input class and a parameter data prediction data class, and specify that the fault detection-related data columns in the time series database are loaded in the first file format, the data columns including the acquisition time column, vibration parameter column and temperature parameter column; Initialize the MLContext object to create an ML.NET environment that is shared among machine learning model creation workflow objects; The data column is converted into an IDataView object, and the SR-CNN algorithm is executed to analyze and label the abnormal data.

5. The machine learning-based motor fault detection method as described in claim 4, characterized in that, The first file is in CSV format.

6. A machine learning-based motor fault detection method, characterized in that, Includes the following steps: S0. Configure the motor model and PLC communication information in the configuration database; S1. Obtain the motor barcode entered by the motor barcode entry device through the host computer; S2. Obtain motor model information through the motor barcode; S3. Read the vibration and temperature data of the motor from the PLC; S4. Store the collected time, motor model information, vibration data, and temperature data into the time series database; S5. First, read the acquisition time and vibration data of the current motor model from the time series database and generate a CSV file; S6. Load the CSV file into the motor fault detection software; S7. Define the vibration data input class and the vibration data prediction data class, and use the LoadColumn attribute to specify the time column and vibration data column of the CSV file to be loaded; S8. Initialize the MLContext object, create the ML.NET environment, and share this environment among the model creation workflow objects; S9. Load the data obtained in step S7 into memory and convert it into an IDataView object; S10. Use the DetectSeasonality function to determine the period value of the timing sequence; S11. Use the DetectEntireAnomalyBySrCnn method to find and label abnormal vibration data, and generate a motor vibration fault warning. S12. First, read the acquisition time and temperature data of the current motor model from the time series database and generate a CSV file. Repeat the above steps S6~S11 to find abnormal temperature data and mark them to generate a motor temperature fault warning.

Images