Motor fault detection method and detection equipment

Abstract

The invention discloses a motor fault detection method and detection equipment, and belongs to the technical field of motors, and the method comprises the steps: obtaining M-dimensional first sensing data of M sensors for a motor; preprocessing the M-dimensional first sensing data to obtain M-dimensional second sensing data; performing feature extraction on the M-dimensional second sensing data to obtain N-dimensional motor features; and inputting the N-dimensional motor features into a trained fault prediction model, carrying out feature fusion on the N-dimensional motor features through the fault prediction model, and then carrying out classification prediction to obtain a fault detection result of the motor, the fault detection result comprising whether each part of the motor has a fault and the fault type of each part. According to the invention, the technical problem of high false alarm rate and missing report rate of motor fault diagnosis is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of motors, and in particular relates to a motor fault detection method and detection equipment. Background Art

[0002] As the core power equipment of industrial production, the operating status of motors directly affects production efficiency, equipment life and safety. The significance of motor status detection and fault diagnosis is not only reflected in the technical level, but is also closely related to economic benefits, safety, environmental protection and industrial intelligent transformation. It can not only prevent sudden failures, but also, by real-time monitoring of parameters such as motor vibration, it can identify potential failures such as bearing wear in advance, avoiding sudden shutdowns that lead to production line interruptions. Especially for steel companies, vibration analysis can provide early warning of bearing failures such as rolling mill motors, avoiding tens of millions of production capacity losses. At the same time, through motor status detection and fault diagnosis, early faults can be accurately diagnosed and targeted maintenance (such as lubrication adjustment or partial replacement of parts, etc.) can be implemented to reduce overall equipment losses.

[0003] In the related art, motor status detection and fault diagnosis rely on a single sensor for detection, which can only detect a certain type of fault (such as a vibration sensor monitoring mechanical faults, but unable to capture electrical faults such as winding short circuits or insulation aging), resulting in limited coverage of fault diagnosis. The installation position of a single sensor is fixed, and there are spatial blind spots, which can easily miss local anomalies (such as undetected bearing wear at the other end of the motor). It is easily affected by environmental noise (such as factory vibration noise leading to misjudgment), and cannot be denoised through cross-validation of multi-source data. Relying on simple threshold judgments (such as temperature over-limit alarms), it is difficult to distinguish between normal fluctuations and early faults, resulting in frequent false alarms or delayed diagnosis. The motor status detection and fault diagnosis in the related art have high false alarm and missed alarm rates. Summary of the Invention

[0004] The embodiments of the present invention provide a motor fault detection method and a detection device to solve the technical problem of high false alarm rate and missed alarm rate in motor state detection and fault diagnosis in the related art.

[0005] According to a first aspect of the present invention, a motor fault detection method is provided, comprising:

[0006] Acquiring M-dimensional first sensing data of the motor from M sensors, the M sensors including an acceleration sensor, a voltage sensor, and at least one temperature sensor; the M-dimensional first sensing data including: a vibration signal of the motor captured by the acceleration sensor, a temperature signal of at least one part of the motor captured by the at least one temperature sensor, and a voltage signal or a current signal of the motor captured by the voltage sensor, where M is an integer greater than 1;

[0007] Preprocessing the M-dimensional first sensor data to obtain M-dimensional second sensor data;

[0008] Performing feature extraction on the M-dimensional second sensor data to obtain N-dimensional motor features, where N is an integer greater than 1;

[0009] The N-dimensional motor features are input into the trained fault prediction model, and the N-dimensional motor features are subjected to feature fusion and classification prediction by the fault prediction model to obtain the fault detection results of the motor. The fault detection results include whether a fault occurs in each part of the motor and the type of fault that occurs in each part.

[0010] In combination with the first aspect, in some embodiments, obtaining M-dimensional first sensing data of the motor from M sensors includes:

[0011] The M sensors are controlled based on a same clock source to collect data from the motor to obtain the M-dimensional first sensing data, where each dimension of the first sensing data is a sampling sequence with a timestamp.

[0012] In combination with the first aspect, in some embodiments: the acceleration sensor is a three-axis acceleration sensor with a sampling frequency in the range of 5-10kHz; the sampling frequency of each temperature sensor is in the range of 1-10Hz and the sampling temperature is in the range of 0-60°C, and the at least one temperature sensor is used to capture the temperature signal of at least one part of the motor winding, motor bearing and motor housing; and the sampling frequency of the voltage sensor is in the range of 1-5kHz, and is used to capture the three-phase voltage signal or three-phase current signal of the motor.

[0013] In combination with the first aspect, in some embodiments, preprocessing the M-dimensional first sensor data to obtain the M-dimensional second sensor data includes:

[0014] performing noise reduction processing on the M-dimensional first sensor data to obtain M-dimensional noise-reduced sensor data, wherein the vibration signal is band-pass filtered, the temperature signal is sliding average filtered, and the voltage signal or the current signal is low-pass filtered;

[0015] Based on the timestamps of the M-dimensional denoised sensor data, interpolation alignment is performed on the M-dimensional denoised sensor data to obtain M-dimensional aligned sensor data, wherein each dimension of the denoised sensor data is a sampling sequence with a timestamp;

[0016] Abnormal sampling points in the M-dimensional aligned sensor data are eliminated to obtain the M-dimensional second sensor data.

[0017] In combination with the first aspect, in some embodiments, feature extraction is performed on the M-dimensional second sensor data to obtain N-dimensional motor features, including:

[0018] Performing time domain analysis, frequency domain analysis, and wavelet transform processing on the vibration signal in the M-dimensional second sensor data, extracting at least one-dimensional motor features as follows through the time domain analysis: the root mean square, peak value, kurtosis, and form factor of the vibration signal in the M-dimensional second sensor data; extracting at least one-dimensional motor features as follows through the frequency domain analysis: the fundamental frequency and harmonic components of the vibration signal in the M-dimensional second sensor data; and extracting at least one-dimensional motor features as follows through the wavelet transform: a transient impact signal of the vibration signal in the M-dimensional second sensor data;

[0019] Extracting at least one dimension of the following motor features from the temperature signal in the M-dimensional second sensor data: mean, variance, maximum value, temperature change rate, steady-state value and temperature gradient of the temperature signal in the M-dimensional second sensor data;

[0020] At least one dimension of the following motor features is extracted from the voltage signal in the M-dimensional second sensor data: voltage imbalance, total harmonic distortion and power factor of the voltage signal in the M-dimensional second sensor data are extracted.

[0021] In combination with the first aspect, in some embodiments, the fault prediction model includes K neural network branches, and a fully connected layer connected to output ends of the K neural network branches, where K is an integer greater than 1;

[0022] The fault prediction model is used to perform feature fusion and classification prediction on the N-dimensional motor features to obtain a fault detection result of the motor, including:

[0023] Based on each of the neural network branches, inputting each dimension of the motor features corresponding to the neural network branch in the N-dimensional motor features into the neural network branch for feature extraction to obtain a feature subspace corresponding to the neural network branch;

[0024] The K-dimensional feature subspace is input into the fully connected layer, and the K-dimensional feature subspace is subjected to feature fusion by the fully connected layer and then classified and predicted to obtain the fault detection result of the motor. The fault detection result includes whether a fault occurs in each part of the bearing, rotor and stator winding of the motor and the fault type of the fault.

[0025] In combination with the first aspect, in some embodiments, inputting each dimension of the N-dimensional motor features corresponding to the neural network branch into the neural network branch for feature extraction to obtain a feature subspace corresponding to the neural network branch includes:

[0026] Inputting each dimension of the N-dimensional motor features related to vibration into a first neural network branch for feature extraction to obtain a first vibration feature subspace;

[0027] Inputting each dimension of the N-dimensional motor features related to voltage into a second neural network branch for feature extraction to obtain a first voltage feature subspace;

[0028] Inputting each dimension of the N-dimensional motor features related to temperature into a third neural network branch for feature extraction to obtain a second vibration feature subspace;

[0029] The motor features of each dimension related to the rotational speed in the N-dimensional motor features are input into the third neural network branch for feature extraction to obtain a second voltage feature subspace.

[0030] In conjunction with the first aspect, in some embodiments, the fault prediction model is pre-trained through the following steps:

[0031] Obtaining a training data set, wherein the training data set includes multiple groups of training samples corresponding to multiple fault types, each group of training samples includes multiple training samples, and each training sample includes an M-dimensional sample value and a label of one fault type among the multiple fault types;

[0032] The K neural network branches and the fully connected layer are jointly trained based on the training data set to obtain the fault prediction model.

[0033] In conjunction with the first aspect, in some embodiments, obtaining each training sample in the training data set includes:

[0034] Controlling the experimental prototype to generate a fault of a target fault type, and obtaining M-dimensional sample values ​​through multiple sensors provided on the experimental prototype, wherein the target fault type belongs to one of the multiple fault types;

[0035] The M-dimensional sample values ​​are labeled based on the target fault type to obtain the training samples.

[0036] According to a second aspect of the present invention, a detection device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the motor fault detection method described in any embodiment of the first aspect is implemented.

[0037] The one or more technical solutions provided by the embodiments of the present invention achieve at least the following technical effects or advantages:

[0038] By installing multiple sensors on the motor, M-dimensional first sensor data is collected from the motor. The M-dimensional first sensor data includes the motor's vibration signal captured by an acceleration sensor, the temperature signal of at least one motor part captured by at least one temperature sensor, and the motor's voltage signal or current signal captured by a voltage sensor. This multi-source sensor data is obtained from the motor, eliminating the spatial blind spots of a single sensor and avoiding inaccurate or distorted fault determination. Furthermore, feature extraction and feature fusion are performed on the M-dimensional first sensor data, enriching the data content and enabling targeted fault diagnosis of various parts of the motor. This allows accurate diagnosis of various motor faults and reduces the false alarm and missed alarm rates of motor fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0040] Figure 1 A flow chart showing a motor fault detection method in some embodiments of the present invention is shown;

[0041] Figure 2 Shown Figure 1 Schematic diagram of the fault prediction model;

[0042] Figure 3 A schematic structural diagram of a detection device in some embodiments of the present invention is shown. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Figure 1 FIG. 1 shows a flow chart of a motor fault detection method in some embodiments of the present invention. Figure 1 As shown, the motor fault detection method includes steps S101 to S104.

[0045] In step S101: obtain M-dimensional first sensing data of the motor from M sensors, where the M sensors include an acceleration sensor, a voltage sensor and at least one temperature sensor; the M-dimensional first sensing data includes a vibration signal of the motor captured by the acceleration sensor, a temperature signal of at least one part of the motor captured by at least one temperature sensor, and a voltage signal or current signal of the motor captured by the voltage sensor, where M is an integer greater than 1.

[0046] In some embodiments, the M sensors in step S101 also include a sensor for capturing the rotational speed of a motor, and the M-dimensional first sensing data also include a rotational speed signal of the motor.

[0047] It can be understood that the at least one temperature sensor in step S101 is used to capture the temperature signal of at least one part of the stator winding, bearing and housing of the motor.

[0048] In some embodiments, the temperature sensors include multiple ones, which are arranged at different parts of the motor. The multiple temperature sensors are used to detect temperature signals of different parts of the motor, thereby avoiding missing local temperature anomalies of the motor.

[0049] In some embodiments, the multiple temperature sensors may include: at least one temperature sensor arranged at the stator winding, at least one temperature sensor arranged at the bearing, and at least one temperature sensor arranged at the casing, for correspondingly monitoring the temperature signals at the stator winding, bearings and casing of the motor.

[0050] In some embodiments, the sampling frequency of each temperature sensor is within the range of 1-10 Hz and the sampling temperature is within the range of 0-60° C. The temperature sensor may be a PT100 or a thermocouple, etc., and the type of temperature sensor may be selected according to the actual working environment and temperature of the motor.

[0051] In some embodiments, the acceleration sensor is a three-axis acceleration sensor with a sampling frequency in the range of 5-10 kHz, which is used to capture the vibration signal of the motor, which is related to the rotor, voltage imbalance, bearing wear, etc.

[0052] In some embodiments, the sampling frequency of the voltage sensor is in the range of 1-5kHz, and the voltage sensor is used to capture the three-phase voltage signal or three-phase current signal of the motor. The three-phase voltage signal or three-phase current signal is used to judge voltage fluctuations, phase loss or harmonic distortion, etc.

[0053] In order to provide a basis for subsequent feature fusion and improve the accuracy of motor fault prediction, in some embodiments, M sensors can be controlled based on the same clock source to collect data from the motor to obtain M-dimensional first sensor data, and the obtained first sensor data of each dimension is a sampling sequence with a timestamp. The sampling sequence corresponding to each sensor includes multiple sampling points, and each sampling point is a sampling value with a timestamp obtained by the corresponding sensor once. For example, the temperature signal detected by each temperature sensor on the motor is a temperature sequence with a timestamp. The vibration signal detected by each acceleration sensor on the motor is an acceleration sequence with a timestamp, the voltage signal detected by each voltage sensor on the motor is a voltage sequence with a timestamp, and the current signal detected on the motor is a current sequence with a timestamp.

[0054] In some embodiments, controlling M sensors to collect data from a motor based on a common clock source can be hardware-triggered to ensure time alignment of the M-dimensional first sensor data collected by the M sensors. For sensors that support hardware triggering, an FPGA (Field Programmable Gate Array) can generate a precise trigger signal, such as a pulse signal or a PPS (pulse-per-second) signal, to control the M sensors to collect data at the same sampling time, thereby synchronizing the sampling times of the M sensors.

[0055] In other embodiments, M sensors are controlled based on the same clock source to collect data from the motor. The time alignment of the M-dimensional first sensor data collected by the M sensors can be ensured through GPS (Global Positioning System). Specifically, the Global Positioning System (GPS) time is used as a unified reference time to generate a PPS signal, and the M sensors are controlled to collect data at the same sampling time to synchronize the sampling times of the M sensors.

[0056] In step S102 : pre-process the M-dimensional first sensing data to obtain M-dimensional second sensing data.

[0057] In some embodiments, preprocessing the M-dimensional first sensor data includes performing one or more of the following preprocessing steps: performing noise reduction processing on each dimension of the M-dimensional first sensor data, interpolation alignment based on timestamps, and removing outliers.

[0058] In some embodiments, each dimension of the M-dimensional first sensor data may be subjected to noise reduction processing, timestamp-based interpolation alignment, and outlier removal in sequence to obtain the M-dimensional second sensor data, including the following steps 1 to 3:

[0059] Step 1: Perform noise reduction processing on the M-dimensional first sensor data to obtain M-dimensional noise-reduced sensor data.

[0060] In step 1, the vibration signal is bandpass filtered, the temperature signal is sliding average filtered, and the voltage or current signal is low-pass filtered. Using bandpass filtering to process the vibration signal removes high-frequency noise and low-frequency interference, reducing the impact of environmental noise. Using sliding average filtering to process the temperature signal eliminates ambient temperature disturbances, and using low-pass filtering to process the voltage or current signal removes high-frequency noise.

[0061] Step 2: Based on the timestamps of the M-dimensional denoised sensor data, the M-dimensional denoised sensor data are interpolated and aligned to obtain M-dimensional aligned sensor data. Each dimension of the denoised sensor data is a sampling sequence with a timestamp.

[0062] It is understood that timestamp-based interpolation alignment can be achieved by selecting a primary sensor as the time reference and performing interpolation processing on the noise-reduced sensor data of each dimension corresponding to other sensors, so that the timestamps of the noise-reduced sensor data of each dimension corresponding to other sensors are aligned with the timestamps of the noise-reduced sensor data of the primary sensor. For example, an accelerometer with a higher sampling frequency can be used as the primary sensor. This allows alignment between low-frequency data (temperature signals) and high-frequency data (vibration signals, voltage signals).

[0063] Step 3: Remove abnormal sampling points in the sensor data after M-dimensional alignment to obtain M-dimensional second sensor data. After interpolation alignment, use the Z-score method to remove abnormal sampling points in the sensor data after M-dimensional alignment.

[0064] In step S103 : feature extraction is performed on the M-dimensional second sensor data to obtain N-dimensional motor features, where N is an integer greater than 1.

[0065] In some embodiments, the vibration signal in the M-dimensional second sensor data, that is, the acceleration sequence in the M-dimensional second sensor data, is subjected to time domain analysis, frequency domain analysis, and wavelet transform processing. The time domain analysis extracts at least one of the following motor features: the root mean square, peak value, kurtosis, and form factor of the vibration signal in the M-dimensional second sensor data. The frequency domain analysis extracts at least one of the following motor features: the fundamental frequency and harmonic components of the vibration signal in the M-dimensional second sensor data. The wavelet transform extracts at least one of the following motor features: the transient impact signal of the vibration signal in the M-dimensional second sensor data. It is understandable that the frequency domain analysis is mainly achieved through FFT (Fast Fourier Transform) spectrum analysis.

[0066] In some embodiments, at least one of the following motor features is extracted from the temperature signal in the M-dimensional second sensor data (that is, the temperature sequence in the M-dimensional second sensor data): mean, variance, maximum value, temperature change rate, steady-state value, and temperature gradient of the temperature signal in the M-dimensional second sensor data;

[0067] In some embodiments, at least one dimension of the following motor features is extracted from the voltage signal in the M-dimensional second sensor data (that is, the voltage sequence in the M-dimensional second sensor data): voltage imbalance, total harmonic distortion and power factor of the voltage signal in the M-dimensional second sensor data are extracted.

[0068] By performing feature extraction on each dimension of the M-dimensional second sensor data and the sensor data of each sensor separately, the data meaning of the sensor data of each dimension is retained, and N-dimensional motor features can be extracted from the M-dimensional second sensor data.

[0069] In step S104: the N-dimensional motor features are input into the trained fault prediction model, and the N-dimensional motor features are subjected to feature fusion and classification prediction by the fault prediction model to obtain the motor fault detection results. The fault detection results include whether each part of the motor has a fault and the type of fault that has occurred in each part.

[0070] In some embodiments, the fault prediction model includes K neural network branches and a fully connected layer connected to the output ends of the K neural network branches, where K is an integer greater than 1. In some embodiments, the number of neural network branches can be the same as the number of sensors, that is, one neural network branch is provided for each sensor. In other embodiments, the number of neural network branches can be the same as the type of sensors, that is, one neural network branch is provided for each type of sensor.

[0071] It can be understood that the neural network branch corresponding to processing acceleration sequence adopts CNN (Convolutional Neural Networks) neural network, the neural network branch corresponding to processing temperature sequence adopts LSTM (Long Short-Term Memory), and the neural network branch corresponding to processing voltage sequence adopts a fully connected network.

[0072] In some embodiments, when there is a sensor for detecting the rotational speed of the motor, the K neural network branches may also include a neural network branch corresponding to processing the rotational speed sequence.

[0073] In some embodiments, a fault prediction model is used to perform feature fusion and classification prediction on N-dimensional motor features to obtain a motor fault detection result, including: based on each neural network branch, inputting the motor features of each dimension corresponding to the neural network branch in the N-dimensional motor features into the neural network branch for feature extraction to obtain a feature subspace corresponding to the neural network branch, thereby obtaining a K-dimensional feature subspace based on the one-to-one correspondence of K neural network branches. The K-dimensional feature subspace is input into a fully connected layer, and the K-dimensional feature subspace is subjected to feature fusion and classification prediction by the fully connected layer to obtain a motor fault detection result. The obtained fault detection result includes whether a fault occurs in each part of the motor's bearings, rotor, and stator windings, as well as the fault type of the fault that occurs.

[0074] It is understandable that in the fault detection results, the possible types of bearing faults include: damage to the inner ring, outer ring, rolling elements, etc.; the possible types of rotor faults include: rotor imbalance, misalignment, etc.; the possible types of stator winding faults include: short circuit, insulation aging, voltage phase loss, voltage imbalance, stator winding overheating (lubrication failure, overload), etc.

[0075] For bearing fault detection, the fault prediction model extracts features from vibration signals and temperature signals to obtain an envelope spectrum, extracts the characteristic frequency of the bearing from the envelope spectrum, and obtains the temperature rise feature from the temperature signal. The bearing is judged to be damaged based on the characteristic frequency and temperature rise features in the fused features, and the bearing fault detection result is obtained. For stator winding fault detection, the fault prediction model predicts the insulation aging risk of the stator winding through the characteristics of continuous temperature rise and increased voltage harmonics, and determines whether there is a phase loss through the characteristics of voltage imbalance and whether the low-frequency component in the harmonic component suddenly increases.

[0076] Figure 2 Shown Figure 1 Schematic diagram of the fault prediction model. Figure 2As shown, in some embodiments, each dimension of the motor features corresponding to the neural network branch in the N-dimensional motor features is input into the neural network branch for feature extraction to obtain a feature subspace corresponding to the neural network branch, including: inputting each dimension of the motor features related to vibration in the N-dimensional motor features into the first neural network branch for feature extraction to obtain a first vibration feature subspace; inputting each dimension of the motor features related to voltage in the N-dimensional motor features into the second neural network branch for feature extraction to obtain a first voltage feature subspace; since temperature abnormality may be caused by vibration abnormality and speed abnormality may be caused by voltage abnormality, therefore, each dimension of the motor features related to temperature in the N-dimensional motor features is input into the third neural network branch for feature extraction to obtain a second vibration feature subspace, and each dimension of the motor features related to speed in the N-dimensional motor features is input into the third neural network branch for feature extraction to obtain a second voltage feature subspace.

[0077] In some embodiments, the fault prediction model is pre-trained through the following steps: obtaining a training data set, the training data set including multiple training samples, and the types of fault type labels include: bearing inner ring damage, bearing outer ring damage, bearing rolling element damage, rotor imbalance, rotor misalignment, stator winding short circuit, stator winding insulation aging, voltage phase loss, voltage imbalance, stator winding overheating (lubrication failure, overload), etc. Each training sample has a fault type label, and the training data set covers all the above fault type labels. Increasing the label diversity of training samples can enable the trained fault prediction model to predict various fault types, thereby making the fault detection coverage of the fault prediction model more comprehensive and the detection more accurate.

[0078] It can be understood that in the training data set, each training sample includes M-dimensional sample values ​​and corresponding fault type labels; based on the training data set, K neural network branches and the fully connected layer are jointly trained to obtain a fault prediction model.

[0079] In some embodiments, the steps for obtaining each training sample in the training data set include: controlling an experimental prototype to sequentially generate a fault of one of multiple fault types; obtaining M-dimensional sample values ​​corresponding to the target fault type, using M sensors installed on the experimental prototype, when the experimental prototype generates a target fault type; and labeling the M-dimensional sample values ​​based on the target fault type to obtain a training sample corresponding to the target fault type. This allows different training samples with different fault type labels to be obtained, enriching the training data set.

[0080] According to an embodiment of the present invention, a multimodal fusion network is used to construct a fault prediction model, and the samples used for training have multiple fault types. Thus, intelligent diagnosis of various motor faults such as bearings, winding overheating, and phase loss can be achieved, and the detection of various motor faults can be covered, effectively ensuring the stability and reliability of equipment operation.

[0081] Based on the same inventive concept, the embodiment of the present invention further provides a detection device, such as Figure 3 As shown, the detection device includes a memory 304, a processor 302, and a computer program stored in the memory 304 and executable on the processor 302. The processor 302 executes the program to implement the steps of the above-mentioned motor fault detection method.

[0082] Among them, Figure 3 In the embodiment of the present invention, a bus architecture (represented by bus 300) is shown. Bus 300 may include any number of interconnected buses and bridges, and bus 300 links together various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same component, namely a transceiver, which provides a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 may be used to store data used by processor 302 when performing operations.

[0083] According to the motor fault detection method and detection device provided by the embodiments of the present invention, multiple sensors are provided on the motor to collect M-dimensional first sensor data of the motor. The M-dimensional first sensor data includes the vibration signal of the motor captured by an acceleration sensor, the temperature signal of at least one part of the motor captured by at least one temperature sensor, and the voltage signal or current signal of the motor captured by a voltage sensor, thereby obtaining multi-source sensor data of the motor, solving the spatial blind spot of a single sensor and avoiding the occurrence of inaccurate or distorted fault judgments. On this basis, by extracting and fusing features from the M-dimensional first sensor data, the data content is enriched and used for targeted fault diagnosis of various parts of the motor, thereby accurately diagnosing various motor faults and reducing the false alarm rate and missed alarm rate of motor fault detection.

[0084] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that performs the functions specified in one or more boxes.

[0085] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0087] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0088] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0089] The foregoing description is merely an embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of the claims.

Claims

1. A motor fault detection method, characterized in that: include: Acquire M-dimensional first sensing data of the motor from M sensors, where the M sensors include an acceleration sensor, a voltage sensor, and at least one temperature sensor; The M-dimensional first sensor data includes: a vibration signal of the motor captured by the acceleration sensor, a temperature signal of at least one part of the motor captured by the at least one temperature sensor, and a voltage signal or a current signal of the motor captured by the voltage sensor, where M is an integer greater than 1; Preprocessing the M-dimensional first sensor data to obtain M-dimensional second sensor data; Performing feature extraction on the M-dimensional second sensor data to obtain N-dimensional motor features, where N is an integer greater than 1; The N-dimensional motor features are input into the trained fault prediction model, and the N-dimensional motor features are subjected to feature fusion and classification prediction by the fault prediction model to obtain the fault detection results of the motor. The fault detection results include whether a fault occurs in each part of the motor and the type of fault that occurs in each part.

2. The motor fault detection method according to claim 1, wherein: The step of obtaining M-dimensional first sensing data of the motor from the M sensors includes: The M sensors are controlled based on a same clock source to collect data from the motor to obtain the M-dimensional first sensing data, where each dimension of the first sensing data is a sampling sequence with a timestamp.

3. The motor fault detection method according to claim 2, wherein: The acceleration sensor is a three-axis acceleration sensor with a sampling frequency in the range of 5-10kHz; The sampling frequency of each temperature sensor is within the range of 1-10 Hz and the sampling temperature is within the range of 0-60° C., and the at least one temperature sensor is used to capture the temperature signal of at least one part of the motor winding, the motor bearing and the motor housing; as well as The sampling frequency of the voltage sensor is in the range of 1-5 kHz, and is used to capture the three-phase voltage signal or three-phase current signal of the motor.

4. The motor fault detection method according to claim 3, wherein: The preprocessing of the M-dimensional first sensor data to obtain the M-dimensional second sensor data includes: performing noise reduction processing on the M-dimensional first sensor data to obtain M-dimensional noise-reduced sensor data, wherein the vibration signal is band-pass filtered, the temperature signal is sliding average filtered, and the voltage signal or the current signal is low-pass filtered; Based on the timestamps of the M-dimensional denoised sensor data, interpolation alignment is performed on the M-dimensional denoised sensor data to obtain M-dimensional aligned sensor data, wherein each dimension of the denoised sensor data is a sampling sequence with a timestamp; Abnormal sampling points in the M-dimensional aligned sensor data are eliminated to obtain the M-dimensional second sensor data.

5. The motor fault detection method according to claim 1, wherein: Feature extraction is performed on the M-dimensional second sensor data to obtain N-dimensional motor features, including: Performing time domain analysis, frequency domain analysis, and wavelet transform processing on the vibration signal in the M-dimensional second sensor data, extracting at least one-dimensional motor features as follows through the time domain analysis: the root mean square, peak value, kurtosis, and form factor of the vibration signal in the M-dimensional second sensor data; extracting at least one-dimensional motor features as follows through the frequency domain analysis: the fundamental frequency and harmonic components of the vibration signal in the M-dimensional second sensor data; and extracting at least one-dimensional motor features as follows through the wavelet transform: a transient impact signal of the vibration signal in the M-dimensional second sensor data; Extracting at least one dimension of the following motor features from the temperature signal in the M-dimensional second sensor data: mean, variance, maximum value, temperature change rate, steady-state value and temperature gradient of the temperature signal in the M-dimensional second sensor data; At least one dimension of the following motor features is extracted from the voltage signal in the M-dimensional second sensor data: voltage imbalance, total harmonic distortion and power factor of the voltage signal in the M-dimensional second sensor data are extracted.

6. The motor fault detection method according to claim 1, wherein: The fault prediction model includes K neural network branches and a fully connected layer connected to the output ends of the K neural network branches, where K is an integer greater than 1; The fault prediction model is used to perform feature fusion and classification prediction on the N-dimensional motor features to obtain a fault detection result of the motor, including: Based on each of the neural network branches, inputting each dimension of the motor features corresponding to the neural network branch in the N-dimensional motor features into the neural network branch for feature extraction to obtain a feature subspace corresponding to the neural network branch; The K-dimensional feature subspace is input into the fully connected layer, and the K-dimensional feature subspace is subjected to feature fusion by the fully connected layer and then classified and predicted to obtain the fault detection result of the motor. The fault detection result includes whether a fault occurs in each part of the bearing, rotor and stator winding of the motor and the fault type of the fault.

7. The motor fault detection method according to claim 6, wherein: The step of inputting each dimension of the N-dimensional motor features corresponding to the neural network branch into the neural network branch for feature extraction to obtain a feature subspace corresponding to the neural network branch includes: Inputting each dimension of the N-dimensional motor features related to vibration into a first neural network branch for feature extraction to obtain a first vibration feature subspace; Inputting each dimension of the N-dimensional motor features related to voltage into a second neural network branch for feature extraction to obtain a first voltage feature subspace; Inputting each dimension of the N-dimensional motor features related to temperature into a third neural network branch for feature extraction to obtain a second vibration feature subspace; The motor features of each dimension related to the rotational speed in the N-dimensional motor features are input into the third neural network branch for feature extraction to obtain a second voltage feature subspace.

8. The motor fault detection method according to claim 6, wherein: The fault prediction model is pre-trained through the following steps: Obtaining a training data set, wherein the training data set includes multiple groups of training samples corresponding to multiple fault types, each group of training samples includes multiple training samples, and each training sample includes an M-dimensional sample value and a label of one fault type among the multiple fault types; The K neural network branches and the fully connected layer are jointly trained based on the training data set to obtain the fault prediction model.

9. The motor fault detection method according to claim 8, wherein: Acquisition of each training sample in the training data set includes: Controlling the experimental prototype to generate a fault of a target fault type, and obtaining M-dimensional sample values ​​through multiple sensors provided on the experimental prototype, wherein the target fault type belongs to one of the multiple fault types; The M-dimensional sample values ​​are labeled based on the target fault type to obtain the training samples.

10. A detection device, characterized in that: include: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the motor fault detection method according to any one of claims 1 to 9 is implemented.

Images