A motor fault diagnosis method based on voiceprint feature extraction and transfer learning

Abstract

The application discloses a motor fault diagnosis method based on voiceprint feature extraction and transfer learning, and belongs to the technical field of equipment fault diagnosis, and comprises the following steps: S1, acquiring a sound signal of a motor working state, and extracting a voiceprint feature through an empirical mode decomposition and a mel cepstrum coefficient method; S2, constructing a voiceprint feature dataset according to the extracted voiceprint feature; and S3, inputting the voiceprint feature dataset into a CNN+ResNet network to obtain a motor fault diagnosis result. The application can overcome the limitations of a traditional contact method, reduce maintenance costs, and also provide more accurate diagnosis information in an early stage of motor fault, thereby improving the operation safety and reliability of motor equipment. Through voiceprint feature extraction technology and data analysis and fault classification of a deep learning model, a new direction is opened up for motor fault diagnosis, and the application has important research value and application prospect.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of equipment fault diagnosis, and particularly relates to a motor fault diagnosis method based on voiceprint feature extraction and transfer learning. BACKGROUND

[0002] With the increasing maturity of deep learning technology, using deep learning methods for motor fault diagnosis has attracted widespread attention, especially in dealing with complex nonlinear and high-dimensional data. Compared with traditional methods, deep learning can automatically learn motor fault features through a large amount of data, and has strong adaptability and generalization ability.

[0003] However, although deep learning has shown great potential and advantages in motor fault diagnosis, existing diagnosis methods based on contact data acquisition still have some significant limitations. For example, the deep learning model training process needs to rely on a large amount of fault data, and current deep learning-based motor fault diagnosis methods usually rely on obtaining motor working state data through contact sensors (such as vibration sensors, accelerometers, current sensors, etc.). These data include vibration signals, high-frequency current signals, SPM pulse currents, etc. of the motor in healthy or fault states, which constitute the data set for training the deep learning model. The core of these methods is to analyze the complex data collected by the sensors through the deep learning model, automatically extract the features of the motor fault, and further perform fault diagnosis.

[0004] However, this approach relying on contact sensors has certain limitations and challenges:

[0005] (1) Need to attach sensors, cannot achieve non-contact detection: Existing fault diagnosis methods generally rely on directly installing sensors on the surface or inside the motor, which not only requires physical contact with the motor, but also requires frequent maintenance and calibration. This approach limits its application in high-risk or difficult-to-contact environments, and in some cases may have some impact on the operation of the motor itself.

[0006] (2) High diagnosis cost: Contact sensors usually need to be installed at key positions of the motor, and different types of faults may require different sensors to collect multiple data, which greatly increases the installation and maintenance cost of the equipment. In addition, the sensor itself also has a certain price, and is easily affected by environmental factors (such as temperature, humidity, vibration, etc.), resulting in instability and inaccuracy of the data, further increasing the cost of the system.

[0007] (3) The installation position is sensitive to the fault diagnosis result: the installation position of the sensor, the selection of the sensor and its configuration and other factors can greatly affect the final diagnosis result. And sensitive to environmental interference, in some extreme working conditions (such as high temperature, high humidity, strong vibration, etc.), the stability and reliability of the sensor may be affected, resulting in reduced accuracy of data acquisition, thereby affecting the reliability of the diagnosis result. SUMMARY

[0008] In view of the above problems in the prior art, the motor fault diagnosis method based on voiceprint feature extraction and transfer learning provided by the present application solves the problems of the prior art method using a contact sensor, such as the need to attach a sensor, the inability to achieve non-contact detection, high diagnosis cost, and the sensitivity of the installation position to the fault diagnosis result.

[0009] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: a motor fault diagnosis method based on voiceprint feature extraction and transfer learning, comprising the following steps:

[0010] S1, obtaining the sound signal of the motor working state, extracting the voiceprint feature by empirical mode decomposition and mel-frequency cepstral coefficient method;

[0011] S2, constructing a voiceprint feature dataset according to the extracted voiceprint feature;

[0012] S3, inputting the voiceprint feature dataset into the CNN+ResNet network to obtain the motor fault diagnosis result.

[0013] Further, the S1 comprises the following steps:

[0014] S11, performing empirical mode decomposition on the sound signal to obtain a plurality of intrinsic mode components, and selecting the three intrinsic mode components with the highest energy contribution degree as the main components of the sound signal;

[0015] S12, calculating the mel-frequency cepstral coefficient of each intrinsic mode component to obtain the cepstral feature of the intrinsic mode component in the mel-frequency band, and combining the mel-frequency cepstral coefficient of the main component to generate the voiceprint feature, wherein the voiceprint feature includes multi-scale and multi-time-frequency feature information.

[0016] Further, the S2 is specifically:

[0017] According to the voiceprint feature, a three-channel dataset is constructed as the voiceprint feature dataset.

[0018] Further, the S3 comprises the following steps:

[0019] S31, inputting the voiceprint feature dataset into the CNN model to perform one-stage feature extraction to obtain the time-frequency feature information of each channel voiceprint feature;

[0020] S32, input the time-frequency feature information and the voiceprint feature dataset into a pre-trained ResNet network, perform two-stage feature extraction, and obtain motor fault features;

[0021] S33, performing feature mapping according to the motor fault features to obtain a motor fault diagnosis result.

[0022] Further, the S32 is specifically:

[0023] The time-frequency feature information is input into the pre-trained ResNet network for training, according to the training result and the task similarity between the voiceprint feature dataset and the image field, the ResNet front-end parameters are frozen, the end parameters are fine-tuned using the voiceprint feature dataset, so that they adapt to the feature scene of motor fault diagnosis, and the motor fault features are output.

[0024] The motor fault diagnosis method based on voiceprint feature extraction and transfer learning is provided, the voiceprint feature extraction method fusing empirical mode decomposition and mel-frequency cepstral coefficient is proposed to construct a voiceprint feature dataset, then the pre-training features of transfer learning and ResNet are fully utilized to realize motor fault prediction and diagnosis under a small amount of voiceprint feature data scene, compared with the prior art, the following advantages are obtained:

[0025] (1) Non-contact and remote monitoring: the sound signal can be remotely collected through a microphone array, without the need of direct contact with the motor, which can effectively avoid the complexity of installation and maintenance of contact sensors, especially in high temperature, high pressure or other dangerous environments, the motor can still be effectively monitored and diagnosed.

[0026] (2) Low diagnosis cost: the sound signal under the working state of the motor can be collected through the microphone array, and the hardware cost is lower than that of the vibration sensor, high-frequency current transformer and the like, and even the microphone of a mobile phone or other handheld terminal can also collect the sound signal for fault diagnosis, with low diagnosis cost.

[0027] (3) Strong environmental adaptability: in some special working environments (such as environments with strong electromagnetic interference, environments with extremely high or low temperature), the precision and stability of the vibration sensor may be affected, while the collection of the sound signal is not easily disturbed by these external factors, so in complex environments, the sound signal can still provide reliable information, and the installation position of the attached sensor is avoided to affect the fault diagnosis.

[0028] (4) Real-time performance and efficiency: Acoustic signal acquisition and processing are relatively simple, and real-time monitoring and analysis can be performed during operation to promptly detect potential motor faults and ensure equipment safety and efficiency. In particular, by combining the CNN+ResNet two-stage transfer learning fault diagnosis strategy, the reliance on fault data can be greatly reduced based on the pre-training results of ResNet, realizing a lightweight motor fault diagnosis scheme, obtaining motor fault diagnosis results more quickly, and achieving more effective fault handling and decision-making basis.

[0029] In summary, the motor fault diagnosis scheme based on voiceprint feature extraction and transfer learning proposed in this invention overcomes the limitations of traditional contact-based methods, reduces maintenance costs, and provides more accurate diagnostic information in the early stages of motor faults, thereby improving the operational safety and reliability of motor equipment. By combining voiceprint feature extraction technology with deep learning models for data analysis and fault classification, this approach opens up new directions for motor fault diagnosis and has significant research value and application prospects. Attached Figure Description

[0030] Figure 1 This is a flowchart of a motor fault diagnosis method based on voiceprint feature extraction and transfer learning according to the present invention.

[0031] Figure 2 This is a schematic diagram of the overall technical roadmap.

[0032] Figure 3 This describes the workflow for voiceprint feature extraction based on EMD and MFCC.

[0033] Figure 4 The process of constructing a voiceprint dataset.

[0034] Figure 5 This describes a two-stage transfer learning fault diagnosis strategy based on CNN+ResNet. Detailed Implementation

[0035] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0036] like Figure 1 As shown, in one embodiment of the present invention, a motor fault diagnosis method based on voiceprint feature extraction and transfer learning includes the following steps:

[0037] S1, an acoustic signal of a motor working state is acquired, and an acoustic feature is extracted through empirical mode decomposition and a mel-frequency cepstral coefficient method;

[0038] S2, an acoustic feature dataset is constructed according to the extracted acoustic feature;

[0039] S3, the acoustic feature dataset is input into a CNN+ResNet network, and a motor fault diagnosis result is obtained.

[0040] In the embodiment, the application provides a motor fault diagnosis method based on acoustic feature extraction and transfer learning, and the overall scheme architecture is as shown in Figure 2 The method mainly includes three execution steps: (1) acoustic feature extraction based on EMD and MFCC; (2) construction of an acoustic feature dataset according to MFCC of each intrinsic mode component (IMF); and (3) construction of a two-stage transfer learning fault diagnosis strategy based on CNN+ResNet.

[0041] The S1 includes the following sub-steps:

[0042] S11, the acoustic signal is subjected to empirical mode decomposition to obtain a plurality of intrinsic mode components, and the three intrinsic mode components with the highest energy contribution degree (Ecd) are selected as main components of the acoustic signal;

[0043] S12, the mel-frequency cepstral coefficient of each intrinsic mode component is calculated to obtain the cepstral feature of the intrinsic mode component on the mel-frequency band, and the mel-frequency cepstral coefficient of the main component is combined to generate an acoustic feature, and the acoustic feature includes multi-scale and multi-time-frequency feature information.

[0044] As shown in Figure 3 In the embodiment, the application provides an acoustic feature extraction method based on empirical mode decomposition (EMD) and mel-frequency cepstral coefficient (MFCC). The acoustic signal is subjected to empirical mode decomposition to obtain each intrinsic mode component, the energy contribution degree of all intrinsic mode components is calculated according to a formula, the intrinsic mode components are arranged in descending order according to the energy contribution degree, and the first three intrinsic mode components arranged in descending order are selected as the main components of the acoustic signal. The expression of the energy contribution degree of the first intrinsic mode component is specifically as follows:

[0045]

[0046] In the formula, Ei is the energy of the i-th intrinsic mode component, and E is the total energy of all intrinsic mode components.

[0047] The S2 is specifically as follows:

[0048] ​​​​​According to the voiceprint feature, a three-channel data set is constructed as a voiceprint feature data set.

[0049] As shown in the drawings, Figure 4 In this embodiment, a large number of data sets are needed to construct in the process of motor fault diagnosis by using deep learning. Based on the extracted voiceprint feature information, a three-channel data set similar to a color image is constructed for deep learning model training and motor fault diagnosis.

[0050] The S3 includes the following steps:

[0051] S31, input the voiceprint feature data set into the CNN model to perform one-stage feature extraction to obtain time-frequency feature information of each channel voiceprint feature;

[0052] S32, input the time-frequency feature information and the voiceprint feature data set into the pre-trained ResNet network to perform two-stage feature extraction to obtain motor fault features;

[0053] S33, perform feature mapping according to the motor fault features to obtain a motor fault diagnosis result.

[0054] The S32 is specifically:

[0055] The time-frequency feature information is input into the pre-trained ResNet network for training. According to the training result and the task similarity between the voiceprint feature data set and the image field, the front-end parameters of the ResNet are frozen, the front-end convolutional layer parameters of the ResNet pre-training model are retained, and the updating of these parameters in the training process is avoided. The end parameters are trained and fine-tuned using the voiceprint feature data set to adapt to the feature scene of the motor fault diagnosis, and the motor fault features are output.

[0056] As shown in the drawings, Figure 5 In this embodiment, the application fully integrates the feature extraction advantages of CNN and ResNet, and proposes a two-stage transfer learning fault diagnosis strategy based on CNN+ResNet. The CNN model is used to perform one-stage feature extraction on the three-channel voiceprint feature data to extract time-frequency feature information of each channel voiceprint feature. Relying on the large-scale parameter training result of the ResNet and the task similarity between the three-channel voiceprint feature data and the image field, the front-end parameters of the ResNet are frozen, the end parameters are fine-tuned using the voiceprint feature data set to adapt to the feature scene of the motor fault diagnosis, and the task migration of the motor fault diagnosis is realized.

[0057] In the description of the application, it needs to be understood that the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. In addition, the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implying the number of technical features indicated. Therefore, the features defined by "first", "second", "third" can explicitly or implicitly include one or more of the features.

Claims

1.A motor fault diagnosis method based on voiceprint feature extraction and transfer learning, characterized in that, The method comprises the following steps: S1, acquiring an acoustic signal of a motor working state, and extracting a voiceprint feature through empirical mode decomposition and a mel-frequency cepstral coefficient method; S2, constructing a voiceprint feature dataset according to the extracted voiceprint feature; Specifically, a three-channel dataset is constructed according to the voiceprint feature, and the three-channel dataset is used as the voiceprint feature dataset; S3, inputting the voiceprint feature dataset into a CNN+ResNet network to obtain a motor fault diagnosis result; The S3 comprises the following steps: S31, inputting the voiceprint feature dataset into a CNN model to perform one-stage feature extraction, and obtaining time-frequency feature information of each channel voiceprint feature; S32, inputting the time-frequency feature information and the voiceprint feature dataset into a pre-trained ResNet network to perform two-stage feature extraction, and obtaining a motor fault feature; Specifically, the time-frequency feature information is inputted into the pre-trained ResNet network for training, ResNet front-end parameters are frozen according to a training result and a task similarity between the voiceprint feature dataset and an image field, end parameters are fine-tuned using the voiceprint feature dataset, the end parameters are adapted to a feature scene of the motor fault diagnosis, and the motor fault feature is outputted; S33, performing feature mapping according to the motor fault feature to obtain the motor fault diagnosis result; The S1 comprises the following steps: S11, performing empirical mode decomposition on the acoustic signal to obtain a plurality of intrinsic mode components, and selecting three intrinsic mode components with the highest energy contribution degree as main components of the acoustic signal; S12, calculating a mel-frequency cepstral coefficient of each intrinsic mode component to obtain a cepstral feature of the intrinsic mode component on a mel-frequency band, and generating a voiceprint feature by combining the mel-frequency cepstral coefficients of the main components, wherein the voiceprint feature comprises multi-scale and multi-time-frequency feature information.

Images