Intelligent early warning method for new energy vehicle motor fault combined with voiceprint recognition

Abstract

The application discloses a new energy automobile motor fault intelligent early warning method combined with voiceprint recognition, relates to the technical field of fault early warning, solves the technical problems that the existing fault early warning relies on single modal data, it is difficult to capture the weak signal characteristics of early motor faults, and alarm can be triggered only after the fault develops to a certain extent, and the early warning timeliness is insufficient; the application collects the multi-channel voiceprint signals of the automobile motor; the characteristics of the multi-channel voiceprint signals are identified to obtain voiceprint characteristics; the running state of the motor is analyzed according to the voiceprint characteristics, and a fault early warning signal is generated; early and accurate early warning of the motor fault is realized, and the accuracy and timeliness of the fault early warning are improved.

Description

Technical Field

[0001] This application belongs to the field of fault warning and relates to automotive motor fault warning technology, specifically a new energy vehicle motor fault intelligent warning method that combines voiceprint recognition. Background Technology

[0002] The electric motor in a new energy vehicle is the core component driving the vehicle. Its function is to convert electrical energy into mechanical energy, directly determining the vehicle's performance, range, and energy efficiency. It features high efficiency, environmental friendliness, simplified structure, and rapid dynamic response. Intelligent fault early warning for new energy vehicle motors is an important technical means to ensure vehicle safety and improve maintenance efficiency. It can monitor the motor's operating status in real time and achieve early detection and warning of faults. Combining voiceprint recognition technology for intelligent fault early warning of new energy vehicle motors can significantly improve vehicle safety and reliability. This technology collects the motor's operating sound through a non-invasive microphone and uses AI algorithms to analyze the spectrum and time domain characteristics. It can accurately capture the early, weak abnormal sounds of faults such as bearing wear and winding short circuits, providing warnings weeks earlier than traditional sensors and avoiding serious damage. At the same time, voiceprint recognition has strong resistance to environmental interference, can be fused with vibration and current data to reduce false alarm rates, and has low hardware cost and strong compatibility. It provides an efficient and low-cost health management solution for new energy vehicles.

[0003] In existing technologies, fault warning for motors in new energy vehicles is generally based on electrical parameters, vibration signals, and physical parameters such as temperature and pressure. These parameters include analysis of short-circuit faults in the motor and vibration signals and values ​​of the motor housing. This allows for timely diagnosis of faults when they occur. However, existing fault warning technologies rely on single-mode data, making it difficult to capture the weak signal characteristics of early motor faults. Alarms are often only triggered after the fault has developed to a certain extent, resulting in insufficient timeliness of warnings.

[0004] This application provides an intelligent early warning method for new energy vehicle motor faults that combines voiceprint recognition, in order to solve the above-mentioned technical problems. Summary of the Invention

[0005] This application aims to solve at least one of the technical problems existing in the prior art; to this end, this application proposes an intelligent early warning method for new energy vehicle motor faults that combines voiceprint recognition, in order to solve the technical problem that the fault early warning of the prior art relies on single modal data, which makes it difficult to capture the weak signal characteristics of early motor faults, and often only triggers the alarm after the fault has developed to a certain extent, resulting in insufficient early warning timeliness.

[0006] To achieve the above objectives, the first aspect of this application provides a method for intelligent early warning of motor faults in new energy vehicles that combines voiceprint recognition, including: Acquire multi-channel acoustic fingerprint signals from automotive motors; The characteristics of multi-channel voiceprint signals are identified to obtain voiceprint features; The operating status of the motor is analyzed based on the acoustic characteristics, and a fault warning signal is generated.

[0007] Preferably, the step of identifying the features of the multi-channel voiceprint signal to obtain voiceprint features includes: Retrieve multi-channel acoustic fingerprint signals from the car motor; preprocess the multi-channel acoustic fingerprint signals to obtain the initial acoustic fingerprint signal; Spatial features are extracted by utilizing the phase relationship between multiple channels to obtain the cross power spectral density between channels; spatial feature vectors of the initial voiceprint signal are extracted based on cross-correlation-phase transformation; features of the original voiceprint signal are analyzed by using complex-valued covariance matrix and logarithmic mapping to obtain feature vectors; and the cross power spectral density, spatial feature vectors, and feature vectors are integrated into voiceprint features.

[0008] Preferably, the preprocessing of the multi-channel voiceprint signal to obtain the initial voiceprint signal includes: Retrieve multi-channel acoustic fingerprint signals from the automotive motor; select one channel from multiple channels as a reference channel, calculate the cross-correlation function between the remaining channels and the reference channel, and identify the number of delay samples corresponding to the peak; perform shift compensation on the lagging channels based on the number of delay samples. The expression for the cross-correlation function is: ;in, The cross-correlation value; The reference channel signal sequence; The length of the reference channel signal sequence; The signal sequence of the channel to be calibrated; This represents the time delay value; The DC component of each channel is removed based on the DC bias removal function; the voiceprint signal of each channel is filtered and denoised using a filter to obtain the initial voiceprint signal; the expression of the DC bias removal function is: ;in, To remove the DC signal from the AC signal; The original signal sequence; This is the total length of the signal; This represents the numerical value corresponding to the original signal sequence.

[0009] Preferably, the step of extracting spatial features using the phase relationship between multiple channels to obtain the cross-power spectral density between channels includes: Retrieve the initial voiceprint signals from multiple channels; perform a short-time Fourier transform on the initial voiceprint signals of each channel to obtain a complex spectrum matrix; The expression for the Fourier transform is: ;in, This represents the initial voiceprint signal of the c-th channel; For frame indexing time; Frequency index; For window functions; The cross-power spectral density between the two channels is calculated based on the cross-power analysis function; the expression for the cross-power analysis function is: ;in, for The complex conjugate; Let be the cross-power spectral density between channel i and channel j.

[0010] Preferably, the step of extracting the spatial feature vector of the initial voiceprint signal based on cross-correlation-phase transform includes: The time difference between channels is analyzed based on the cross-correlation-phase transform function; the expression of the cross-correlation-phase transform function is: ; in, For time delay variables; It is the normalized cross spectrum after averaging across all time frames; the expression is: ; This indicates averaging over time frames; This represents the cross-correlation-phase transformation function between channel C and channel 1; Peak positions are extracted as time differences using the cross-correlation-phase transform function, and multiple time differences are integrated to obtain a spatial feature vector. ;in, This represents the time difference between channel C and channel 1.

[0011] Preferably, the step of analyzing the features of the original voiceprint signal using a complex-valued covariance matrix and logarithmic mapping to obtain a feature vector includes: The complex matrix is ​​obtained based on the constructed covariance matrix function analysis; the expression of the covariance matrix function is: ; Use logarithmic mapping to perform eigenvalue decomposition analysis on the complex matrix of each frequency point to obtain eigenvectors; in, Indicates conjugate transpose; This represents the covariance matrix at time frame t and frequency point f. This represents a multi-channel complex spectrum vector at the same time frame t and frequency point f; Representing vectors The conjugate transpose of; The covariance matrix belongs to the complex field with dimension . OK List; This represents standard matrix multiplication.

[0012] Preferably, the step of performing eigenvalue decomposition analysis on the complex matrix of each frequency point using logarithmic mapping to obtain eigenvectors includes: Based on formula For each frequency point, the complex matrix Perform eigenvalue decomposition; where, These are real eigenvalues; It is a unitary matrix; express The i-th element on the diagonal; This is the diagonalization operation function; Based on formula Calculate the logarithm of the matrix to obtain the logarithmic matrix; extract the triangular real and imaginary parts of the logarithmic matrix and arrange them into vectors to obtain the eigenvectors; The expression for the feature vector is: ;in, Represents the real part of a complex number; Represents the imaginary part of a complex number; This represents the element located in the p-th row and q-th column of the logarithmic matrix.

[0013] Preferably, the step of analyzing the motor's operating status based on acoustic signature characteristics and generating a fault warning signal includes: Retrieve voiceprint features within a specified time period; sort and integrate the voiceprint features in chronological order to obtain a state analysis sequence; The process involves calling the state analysis model, inputting the state analysis sequence into the state analysis model to obtain state labels, matching the state labels to obtain the motor's operating state, and generating an early warning signal when the motor's operating state is abnormal. The state analysis model is constructed based on an artificial intelligence model.

[0014] Preferably, the state analysis model is constructed based on an artificial intelligence model, including: Select a model framework and deep learning algorithm from the artificial intelligence library; build a model based on the model framework and deep learning algorithm to obtain the initial model; Obtain the standard dataset; the standard dataset includes standard input data consistent with the content attributes of the state analysis sequence, and standard output data consistent with the content attributes of the state labels; The standard dataset is divided into a training set, a validation set, and a test set according to a set ratio; the initial model is trained using the training set; the internal structure and parameters of the initial model are adjusted using the validation set; and the trained initial model is tested using the test set to obtain test metrics. Obtain the indicator threshold; compare the test indicator with the indicator threshold; if all test indicators are greater than the indicator threshold, mark the initial model as a state analysis model; otherwise, rebuild and retrain the initial model.

[0015] A second aspect of this application provides a computer-readable storage medium storing instructions for performing the methods described in the first aspect and any possible implementation thereof.

[0016] Compared with the prior art, the beneficial effects of this application are: 1. This application provides a comprehensive and effective processing and analysis of multi-channel acoustic signature signals from automotive motors. Firstly, in signal alignment and preliminary processing, cross-correlation functions are calculated to achieve time alignment of each channel, removing DC components and noise, improving signal quality, and providing clean data for subsequent analysis. Next, in frequency domain analysis and feature extraction, short-time Fourier transform is used to convert the signal to the frequency domain, and cross-power spectral density is calculated to reflect channel correlation. Based on the cross-correlation-phase transform function, the time difference between channels is extracted to construct spatial feature vectors, aiding in sound source localization and spatial acoustic analysis. Finally, in matrix analysis and feature vector extraction, a covariance matrix is ​​constructed and eigenvalues ​​are decomposed to reveal the signal's statistical characteristics and correlations. The logarithm of the matrix is ​​calculated, and the real and imaginary parts of the logarithmic matrix are extracted to form feature vectors, which can more effectively capture complex signal features and provide more discriminative features for subsequent classification, recognition, and other tasks.

[0017] 2. This application constructs a complete and scientific automotive motor state analysis system, offering significant benefits. In the model building phase, a suitable model framework and deep learning algorithm are selected from an artificial intelligence library to build an initial model, laying the foundation for subsequent optimization. Standard datasets are used to divide the model into training, validation, and test sets, conducting model training, parameter tuning, and performance evaluation separately. This systematically improves the model's accuracy and generalization ability, ensuring its effective analysis of motor states in different scenarios. By setting indicator thresholds and comparing them with test indicators, model quality is strictly controlled; only models meeting the requirements are marked as state analysis models, avoiding misjudgments. In practical applications, voiceprint features are retrieved, sorted, and integrated into a state analysis sequence. This sequence is input into the model to obtain state labels, which are then matched with the motor's operating state. This allows for rapid and accurate determination of whether the motor is functioning normally. Once an anomaly is detected, a warning signal can be generated promptly, facilitating timely measures to reduce failure losses and ensure the safe and stable operation of the automotive motor. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the overall method steps of this application; Figure 2 This is a schematic diagram of the feature extraction steps in this application; Figure 3 This is a schematic diagram of the fault warning steps in this application. Detailed Implementation

[0020] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] Please see Figure 1 The first aspect of this application provides a method for intelligent early warning of motor faults in new energy vehicles that combines voiceprint recognition, including: S101. Acquire multi-channel acoustic fingerprint signals from the car motor.

[0022] Specifically, the acquisition of acoustic signature signals is achieved through multiple sound sensors (such as microphone arrays) arranged around or on the motor body. In this embodiment, the acquisition targets are not limited to the motor body itself, but also include key vibration transmission components in the motor's operating environment. For example, sensors can be arranged on the surface of the motor housing, at the location of the motor bearing housing, at the motor mounting bracket, or at the connection point of the reducer housing. It should be understood that the sensor placement should be optimized according to the spatial structure and noise source distribution of the actual vehicle model to maximize the signal-to-noise ratio. Through multi-channel acquisition, spatial distribution information of sound sources can be obtained. Compared with single-channel acquisition, it can effectively distinguish between internal motor fault sound sources and external environmental noise, providing rich raw data support for subsequent feature recognition.

[0023] S102. Identify the features of the multi-channel voiceprint signal to obtain voiceprint features.

[0024] Specifically, feature recognition is the process of transforming the original acoustic signature waveform into key characteristic parameters that characterize the motor's operating state. In this embodiment, the feature recognition process encompasses the entire workflow from signal preprocessing to multi-dimensional feature extraction. The original acoustic signature signal contains a large amount of background noise and irrelevant frequency components, making it difficult to extract effective information through direct analysis. Therefore, the feature recognition step first cleans and calibrates the original signal, and then extracts features from multiple dimensions, including the time domain, frequency domain, and time-frequency domain. For example, spatial features reflecting inter-channel correlation, frequency domain features reflecting signal energy distribution, and higher-order features reflecting signal statistical characteristics are extracted. These features together constitute the acoustic signature feature vector, which can sensitively reflect subtle acoustic changes caused by early faults such as motor bearing wear and winding short circuits.

[0025] S103. Analyze the motor's operating status based on the acoustic signature characteristics and generate a fault warning signal.

[0026] Specifically, state analysis is a process of pattern recognition and classification of voiceprint feature vectors based on a pre-built analysis model. In this embodiment, the extracted voiceprint features are input into the state analysis model. The model compares the feature differences between normal and fault states and outputs the current motor operating status label. When the identified status label corresponds to an abnormal state (such as "bearing outer ring wear" or "stator winding inter-turn short circuit"), the system automatically generates a fault warning signal. This warning signal can be transmitted to the vehicle's central control display or a remote monitoring platform to prompt the driver or maintenance personnel to perform inspection. Through the above scheme, this invention utilizes the high sensitivity of voiceprint signals to early faults, combined with multi-channel feature recognition technology, to achieve early warning triggered at the initial stage of a fault, avoiding serious mechanical damage and significantly improving the operational safety and maintenance economy of new energy vehicle motors.

[0027] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, the above S102 can be specifically implemented through the following S201-S205, which are explained in detail below: S201. Retrieve the multi-channel acoustic signature signal of the car motor; select one channel from multiple channels as the reference channel, calculate the cross-correlation function between the remaining channels and the reference channel, and identify the number of delay samples corresponding to the peak; perform shift compensation on the lagging channel based on the number of delay samples.

[0028] The expression for the cross-correlation function is: ;in, The cross-correlation value; The reference channel signal sequence; The length of the reference channel signal sequence; The signal sequence of the channel to be calibrated; This represents the time delay value.

[0029] It should be noted that by calculating different delays The cross-correlation value is calculated, and the value corresponding to the peak cross-correlation value is found. This value represents the number of time delay samples between the two channel signals. Subsequently, the lagging channel is shifted and compensated based on this number of delay samples. This involves removing or adding zeros to the data sequence to eliminate the time asynchrony problem caused by hardware differences in the multi-channel acquisition system, ensuring that the signals of each channel are strictly aligned on the time axis, and laying the foundation for subsequent extraction of phase correlation features.

[0030] S202. Remove the DC component of each channel based on the DC bias removal function; use a filter to filter and denoise the voiceprint signal of each channel to obtain the initial voiceprint signal.

[0031] The expression for the DC bias removal function is: ;in, To remove the DC signal from the AC signal; The original signal sequence; This is the total length of the signal; This represents the numerical value corresponding to the original signal sequence.

[0032] It should be noted that the DC component is usually caused by sensor zero-point drift or circuit bias. Removing the DC component can avoid limiting the dynamic range of the signal and highlight the AC fluctuation component that reflects the motor's operating status. A bandpass filter can be used, and the cutoff frequency can be set according to the frequency range of the motor fault characteristics (such as hundreds of hertz to thousands of hertz) to filter out high-frequency environmental noise (such as wind noise) and low-frequency mechanical vibration interference, thereby outputting an initial acoustic signal with a higher signal-to-noise ratio.

[0033] S203. Perform a short-time Fourier transform on the initial voiceprint signal of each channel to obtain a complex spectral matrix; calculate the cross-power spectral density between the two channels based on the cross-power analysis function.

[0034] The expression for the Fourier transform is: ;in, This represents the initial voiceprint signal of the c-th channel; For frame indexing time; Frequency index; For window functions; The expression for the cross-power analysis function is: ;in, for The complex conjugate; Let be the cross-power spectral density between channel i and channel j.

[0035] It should be noted that the Hanning window is preferred as the window function. The Hanning window has good frequency resolution and low sidelobe leakage characteristics, effectively suppressing spectral leakage and allowing the transformed spectrum to more clearly reflect the distribution of the motor fault characteristic frequencies. It should be understood that the choice of window function is not limited to the Hanning window. In other embodiments, depending on the different trade-offs between time resolution and frequency resolution, a Hamming window or a Gaussian window can also be selected. For example, if a more precise location of the fault occurrence time is required, a window function with higher time resolution can be selected; if a more refined analysis of the frequency components of the fault characteristics is required, a window function with higher frequency resolution can be selected.

[0036] It should be noted that cross-power spectral density not only includes the energy product information of the two channel signals at various frequency points, but more importantly, it preserves the phase difference information. When a local fault occurs inside the motor (such as bearing pitting), the sound waves generated by the fault will reach sensors at different locations at specific angles, resulting in a specific phase difference between the signals of each channel. By calculating the cross-power spectral density, this phase difference information can be effectively extracted, thereby retrieving the spatial location or propagation characteristics of the sound source. Compared to using only the self-power spectral density of a single channel, cross-power spectral density can eliminate background noise common to all channels (such as vehicle wind noise and road noise), because these common-mode noises are usually isotropic or low-correlation in space and will be significantly suppressed in the cross-power spectrum, thus highlighting the fault signal characteristics with a specific spatial propagation path.

[0037] S204. Analyze the time difference between channels based on the cross-correlation-phase transform function; extract the peak position as the time difference based on the cross-correlation-phase transform function, and integrate multiple time differences to obtain a spatial feature vector. .

[0038] The expression for the cross-correlation-phase transform function is: ; in, For time delay variables; It is the normalized cross spectrum after averaging across all time frames; the expression is: ; This indicates averaging over time frames; This represents the cross-correlation-phase transformation function between channel C and channel 1; This represents the time difference between channel C and channel 1.

[0039] It should be noted that in the cross-correlation-phase transform function, normalization is used to eliminate the influence of signal amplitude information, retaining only phase information. In the acoustic signature signal generated by motor faults, the energy differences between different frequency components are huge. If the cross-power spectrum is used directly, the strong frequency components will dominate the peak position of the cross-correlation function, masking the time delay information of weak fault characteristics.

[0040] S205. Based on the constructed covariance matrix, a complex matrix is ​​obtained through functional analysis; based on the formula... For each frequency point, the complex matrix Perform eigenvalue decomposition; based on formula Calculate the logarithm of the matrix to obtain the logarithmic matrix; extract the triangular real and imaginary parts of the logarithmic matrix and arrange them into vectors to obtain the eigenvectors.

[0041] The expression for the covariance matrix function is: ;in, Indicates conjugate transpose; This represents the covariance matrix at time frame t and frequency point f. This represents a multi-channel complex spectrum vector at the same time frame t and frequency point f; Representing vectors The conjugate transpose of; The covariance matrix belongs to the complex field with dimension . OK List; This represents standard matrix multiplication; These are real eigenvalues; It is a unitary matrix; express The i-th element on the diagonal; This is the diagonalization operation function; The expression for the feature vector is: ;in, Represents the real part of a complex number; Represents the imaginary part of a complex number; This represents the element located in the p-th row and q-th column of the logarithmic matrix.

[0042] It should be noted that the constructed complex matrix contains both the signal's self-power spectrum (diagonal elements) and the cross-power spectrum between channels (off-diagonal elements). This means that the matrix not only records the energy distribution of each channel's signal but also fully preserves the phase difference information between channels. For early motor faults (such as bearing microcracks), the resulting acoustic waveform signals often exhibit specific coupling patterns in phase. The complex-valued covariance matrix can sensitively capture these subtle spatial correlation changes, thus providing richer information dimensions for fault identification.

[0043] Based on the above steps, this application provides a comprehensive and effective processing and analysis of multi-channel acoustic signature signals from automotive motors. Firstly, in signal alignment and preliminary processing, cross-correlation functions are calculated to achieve time alignment of each channel, removing DC components and noise, improving signal quality, and providing clean data for subsequent analysis. Next, in frequency domain analysis and feature extraction, short-time Fourier transform is used to convert the signal to the frequency domain, and cross-power spectral density is calculated to reflect channel correlation. Based on the cross-correlation-phase transform function, the time difference between channels is extracted to construct spatial feature vectors, aiding in sound source localization and spatial acoustic analysis. Finally, in matrix analysis and feature vector extraction, a covariance matrix is ​​constructed and eigenvalues ​​are decomposed to reveal the signal's statistical characteristics and correlations. The logarithm of the matrix is ​​calculated, and the real and imaginary parts of the logarithmic matrix are extracted to form feature vectors, which can more effectively capture complex signal features and provide more discriminative features for subsequent classification, recognition, and other tasks.

[0044] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 3As shown, the above S103 can be specifically implemented through the following S301-S306, which are explained in detail below: S301. Select a model framework and deep learning algorithm from the artificial intelligence library; build a model based on the model framework and deep learning algorithm to obtain the initial model.

[0045] It should be noted that the artificial intelligence library includes various mature neural network architectures. Considering the significant temporal dependence of voiceprint signals, this embodiment preferably uses a recurrent neural network (RNN) or a long short-term memory network (LSTM) as the model framework. These models excel at processing sequential data and can effectively capture long-distance dependencies in time series. Of course, in other embodiments, a one-dimensional convolutional neural network (1D-CNN) or a Transformer model can also be selected. CNNs are good at extracting local features, while Transformers have advantages in capturing global correlations. After selecting a suitable model framework, an initial model is constructed by combining it with corresponding deep learning algorithms (such as backpropagation algorithm, Adam optimizer, etc.).

[0046] S302. Obtain the standard dataset; wherein, the standard dataset includes standard input data consistent with the content attributes of the state analysis sequence, and standard output data consistent with the content attributes of the state label.

[0047] It should be noted that the standard input data is the voiceprint feature sequence collected and processed during historical operation, while the standard output data is the actual status label after expert diagnosis or actual disassembly verification. The dataset should cover the normal operation data of the motor under various operating conditions (different speeds, different loads) as well as the operation data under various typical fault modes to ensure that the model has good generalization ability.

[0048] S303. Divide the standard dataset into a training set, a validation set, and a test set according to a set ratio; train the initial model using the training set; adjust the internal structure and parameters of the initial model using the validation set; and test the trained initial model using the test set to obtain test metrics.

[0049] It should be noted that the preferred ratio is 7:2:1, meaning 70% of the data is used as the training set, 20% as the validation set, and 10% as the test set. This ratio can be adjusted based on the total amount of data; for example, cross-validation can be used for small sample sizes. The initial model is trained using the training set, where it continuously adjusts its internal weights to minimize the error between the predicted and true labels. The validation set is used to adjust the internal structure and parameters of the initial model, such as the learning rate, number of network layers, or number of neurons, to prevent overfitting. The trained initial model is then tested using the test set to obtain test metrics. Test metrics are key quantitative parameters for measuring model performance, primarily including accuracy, recall, and precision. Accuracy reflects the proportion of correct overall judgments by the model; recall reflects the model's ability to detect faulty samples. In fault warning scenarios, recall is particularly crucial because the cost of a false negative (classifying a fault as normal) is far higher than a false positive.

[0050] S304. Obtain the indicator threshold; compare the test indicator with the indicator threshold; if all test indicators are greater than the indicator threshold, mark the initial model as the state analysis model; otherwise, rebuild and train the initial model again.

[0051] It should be noted that the indicator thresholds are qualification lines set according to actual application requirements. For example, the accuracy threshold is set at 95%, and the recall threshold at 98%. When all test indicators exceed the indicator thresholds, it indicates that the model has sufficient recognition ability and robustness, and the initial model is marked as a state analysis model and officially put into use. Otherwise, it indicates that the model performance does not meet the standards, and the initial model needs to be rebuilt and trained, for example, by increasing the amount of training data, optimizing the feature extraction algorithm, or adjusting the network structure, until the indicator requirements are met. Through the above rigorous closed-loop process of construction, training, and testing, it is ensured that the state analysis model can accurately identify early and subtle faults in the actual complex vehicle environment, effectively reducing the false alarm rate and false negative rate, and providing reliable technical support for the safe operation of new energy vehicle motors.

[0052] S305. Retrieve voiceprint features within a set time period; sort and integrate the voiceprint features in chronological order to obtain a state analysis sequence.

[0053] S306. Call the state analysis model; input the state analysis sequence into the state analysis model to obtain state labels; obtain the motor's operating state based on state label matching; when the motor's operating state is abnormal, generate an early warning signal.

[0054] Based on the above steps, this application constructs a complete and scientific automotive motor state analysis system, offering significant benefits. In the model building phase, a suitable model framework and deep learning algorithm are selected from an artificial intelligence library to build an initial model, laying the foundation for subsequent optimization. Standard datasets are used to divide the model into training, validation, and test sets, conducting model training, parameter tuning, and performance evaluation separately. This systematically improves the model's accuracy and generalization ability, ensuring its effective analysis of motor states in different scenarios. By setting indicator thresholds and comparing them with test indicators, model quality is strictly controlled; only models meeting the requirements are labeled as state analysis models, avoiding misjudgments. In practical applications, voiceprint features are retrieved, sorted, and integrated into a state analysis sequence. This sequence is input into the model to obtain state labels, which are then matched with the motor's operating state. This allows for rapid and accurate determination of whether the motor is functioning normally. Once an anomaly is detected, a warning signal can be generated promptly, facilitating timely measures to reduce failure losses and ensure the safe and stable operation of the automotive motor.

[0055] To more intuitively demonstrate the practical application effect of the technical solution of the present invention, this embodiment takes the bearing wear fault early warning of a certain type of new energy vehicle drive motor as an example to describe in detail the method described in Embodiment 1 above. It should be understood that this embodiment is only used to explain the present invention and does not constitute a limitation on the scope of protection of the present invention.

[0056] In this application scenario, the object under test is a 150kW permanent magnet synchronous drive motor that has traveled over 30,000 kilometers. The data acquisition hardware employs a circular array of four high-sensitivity MEMS microphones, with a sampling rate set to 48kHz. The microphone array is positioned on the motor housing near the rear bearing housing, an area typically characterized by the shortest propagation path and highest signal-to-noise ratio for bearing fault sound sources. The system automatically triggers a data acquisition task every 10 minutes, with each acquisition lasting 5 seconds, acquiring four channels of raw acoustic signature signals.

[0057] First, step S101 is executed to acquire multi-channel acoustic signature signals from the vehicle's motor. During high-speed cruising, the motor's operating noise is mixed with broadband background noise generated by road surface excitation and high-frequency noise from the vehicle's air conditioning fan. In the acquired raw signals, the weak impact components caused by early bearing wear are almost completely submerged by the background noise, making it difficult to discern any anomalies in the time-domain waveform with the naked eye.

[0058] Subsequently, step S102 is executed to identify the characteristics of the multi-channel voiceprint signal. Specifically, preprocessing is performed first. Due to the difference in line length of each microphone channel, the system uses a cross-correlation function to calculate the time delay between channels. It is found that channel 3 lags behind the reference channel 1 by 2 sampling points, and shift compensation is then performed on this channel to ensure the synchronization of the sound source signal. Next, a DC bias removal function is used to eliminate sensor zero-point drift, and a bandpass filter with a bandwidth of 500Hz to 8000Hz is applied to filter out low-frequency road vibration noise and high-frequency electromagnetic switching noise, resulting in an initial voiceprint signal with a significantly improved signal-to-noise ratio.

[0059] Next, the system extracts three types of features. In cross-power spectral density extraction, STFT transformation reveals a significant energy peak at the second harmonic of the motor's rotational frequency, with stable phase differences between channels, reflecting the mechanical vibration characteristics of the motor. In spatial feature vector extraction, the spatial feature vectors calculated using the GCC-PHAT algorithm show that the time difference between the sound source and different microphones closely matches the bearing housing position, effectively eliminating interference from the roof-mounted air conditioning fan (located in different spaces). In feature vector extraction, the logarithmic mapping feature of the complex-valued covariance matrix keenly captures the non-Gaussian nature of the signal; its eigenvalues ​​exhibit a discrete pattern in the logarithmic domain that differs from the normal state, typically characteristic of transient impact signals caused by early pitting failures in bearings. The system concatenates these three types of feature vectors to form a high-dimensional acoustic signature.

[0060] Finally, step S103 is executed to analyze the motor's operating status based on the acoustic signature characteristics. The system retrieves acoustic signature characteristics generated in the past hour, constructs a state analysis sequence, and inputs it into a pre-trained LSTM state analysis model. The model outputs a state label of 4 corresponding to the state "early wear of the bearing outer ring," with a confidence level of 92%. The system then generates a fault warning signal and sends it to the vehicle controller via the CAN bus, prompting the driver that "the motor bearing is abnormal and maintenance is recommended."

[0061] A second aspect of this application provides a computer-readable storage medium storing instructions for performing the methods described in the first aspect and any possible implementation thereof.

[0062] Specifically, a computer-readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. In this embodiment, a computer-readable storage medium can specifically be a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, magnetic tape, a floppy disk, an optical disk (such as a compact optical disc read-only memory (CD-ROM) or a digital versatile optical disc (DVD), a hard disk, flash memory, a USB flash drive, etc. It should be understood that with the development of storage technology, any new computer-readable storage media that may emerge in the future, as long as they can store the aforementioned instructions and be read by a processor, should be included within the scope of protection of this invention.

[0063] More specifically, the instructions stored in the storage medium are a series of computer-executable codes or commands, which correspond to the various functional modules or steps described in Example 1. When the computer's processor (such as the vehicle controller VCU, motor controller MCU, or independent fault diagnosis chip in a new energy vehicle) reads and executes the instructions in the storage medium, the processor will schedule the corresponding hardware resources according to the logic of the instructions, and sequentially perform operations such as multi-channel acoustic signal acquisition, signal preprocessing (shift compensation, DC removal, filtering), multi-dimensional feature extraction (cross-power spectral density calculation, GCC-PHAT spatial feature vector extraction, complex-valued covariance matrix logarithmic mapping feature extraction), and state analysis model calculation, and finally output a fault warning signal.

[0064] By employing the computer-readable storage medium provided in this embodiment, the aforementioned complex voiceprint recognition and fault early warning algorithms are solidified into storable and distributable program code, greatly reducing the deployment cost and portability difficulty of the technical solution. On the one hand, this medium can be easily integrated into the existing electronic control units of new energy vehicles without large-scale modifications to the hardware architecture; on the other hand, through the copying and transmission of the medium, algorithm upgrades and iterations can be easily achieved, providing solid hardware support for the promotion and application of intelligent early warning technology for motor faults in new energy vehicles.

[0065] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0066] The working principle of this application is as follows: This application collects multi-channel acoustic fingerprint signals from automotive motors; identifies the characteristics of the multi-channel acoustic fingerprint signals to obtain acoustic fingerprint features; analyzes the operating status of the motor based on the acoustic fingerprint features, and generates a fault warning signal.

[0067] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A method for intelligent early warning of motor faults in new energy vehicles combined with voiceprint recognition, characterized in that, include: Acquire multi-channel acoustic fingerprint signals from automotive motors; The characteristics of multi-channel voiceprint signals are identified to obtain voiceprint features; The operating status of the motor is analyzed based on the acoustic characteristics, and a fault warning signal is generated.

2. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 1, characterized in that, The process of identifying the features of multi-channel voiceprint signals to obtain voiceprint features includes: Retrieve multi-channel acoustic fingerprint signals from the car motor; preprocess the multi-channel acoustic fingerprint signals to obtain the initial acoustic fingerprint signal; Spatial features are extracted by utilizing the phase relationship between multiple channels to obtain the cross power spectral density between channels; spatial feature vectors of the initial voiceprint signal are extracted based on cross-correlation-phase transformation; features of the original voiceprint signal are analyzed by using complex-valued covariance matrix and logarithmic mapping to obtain feature vectors; and the cross power spectral density, spatial feature vectors, and feature vectors are integrated into voiceprint features.

3. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 2, characterized in that, The preprocessing of the multi-channel voiceprint signal to obtain the initial voiceprint signal includes: Retrieve multi-channel acoustic fingerprint signals from the automotive motor; select one channel from multiple channels as a reference channel, calculate the cross-correlation function between the remaining channels and the reference channel, and identify the number of delay samples corresponding to the peak; perform shift compensation on the lagging channels based on the number of delay samples. The expression for the cross-correlation function is: ;in, The cross-correlation value; The reference channel signal sequence; The length of the reference channel signal sequence; The signal sequence of the channel to be calibrated; This represents the time delay value; The DC component of each channel is removed based on the DC bias removal function; the voiceprint signal of each channel is filtered and denoised using a filter to obtain the initial voiceprint signal; the expression of the DC bias removal function is: ;in, To remove the DC signal from the AC signal; The original signal sequence; This is the total length of the signal; This represents the numerical value corresponding to the original signal sequence.

4. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 2, characterized in that, The step of extracting spatial features using the phase relationship between multiple channels to obtain the cross-power spectral density between channels includes: Retrieve the initial voiceprint signals from multiple channels; perform a short-time Fourier transform on the initial voiceprint signals of each channel to obtain a complex spectrum matrix; The expression for the Fourier transform is: ;in, This represents the initial voiceprint signal of the c-th channel; For frame indexing time; Frequency index; For window functions; The cross-power spectral density between the two channels is calculated based on the cross-power analysis function; the expression for the cross-power analysis function is: ;in, for The complex conjugate; Let be the cross-power spectral density between channel i and channel j.

5. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 2, characterized in that, The step of extracting the spatial feature vector of the initial acoustic signature signal based on cross-correlation-phase transform includes: The time difference between channels is analyzed based on the cross-correlation-phase transform function; the expression of the cross-correlation-phase transform function is: ; in, For time delay variables; It is the normalized cross spectrum after averaging across all time frames; the expression is: ; This indicates averaging over time frames; This represents the cross-correlation-phase transformation function between channel C and channel 1; Peak positions are extracted as time differences using the cross-correlation-phase transform function, and multiple time differences are integrated to obtain a spatial feature vector. ;in, This represents the time difference between channel C and channel 1.

6. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 2, characterized in that, The method of analyzing the features of the original voiceprint signal using complex-valued covariance matrix and logarithmic mapping yields a feature vector, including: The complex matrix is ​​obtained based on the constructed covariance matrix function analysis; the expression of the covariance matrix function is: ; Use logarithmic mapping to perform eigenvalue decomposition analysis on the complex matrix of each frequency point to obtain eigenvectors; in, Indicates conjugate transpose; This represents the covariance matrix at time frame t and frequency point f. This represents a multi-channel complex spectrum vector at the same time frame t and frequency point f; Representing vectors The conjugate transpose of; The covariance matrix belongs to the complex field with dimension . OK List; This represents standard matrix multiplication.

7. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 6, characterized in that, The method of performing eigenvalue decomposition analysis on the complex matrix of each frequency point using logarithmic mapping to obtain eigenvectors includes: Based on formula For each frequency point, the complex matrix Perform eigenvalue decomposition; where, These are real eigenvalues; It is a unitary matrix; express The i-th element on the diagonal; This is the diagonalization operation function; Based on formula Calculate the logarithm of the matrix to obtain the logarithmic matrix; extract the triangular real and imaginary parts of the logarithmic matrix and arrange them into vectors to obtain the eigenvectors; The expression for the feature vector is: ;in, Represents the real part of a complex number; Represents the imaginary part of a complex number; This represents the element located in the p-th row and q-th column of the logarithmic matrix.

8. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 1, characterized in that, The step of analyzing the motor's operating status based on acoustic signature characteristics and generating a fault warning signal includes: Retrieve voiceprint features within a specified time period; sort and integrate the voiceprint features in chronological order to obtain a state analysis sequence; The process involves calling the state analysis model, inputting the state analysis sequence into the state analysis model to obtain state labels, matching the state labels to obtain the motor's operating state, and generating an early warning signal when the motor's operating state is abnormal. The state analysis model is constructed based on an artificial intelligence model.

9. The intelligent early warning method for new energy vehicle motor faults combined with voiceprint recognition according to claim 8, characterized in that, The state analysis model is built based on an artificial intelligence model and includes: Select a model framework and deep learning algorithm from the artificial intelligence library; build a model based on the model framework and deep learning algorithm to obtain the initial model; Obtain the standard dataset; the standard dataset includes standard input data consistent with the content attributes of the state analysis sequence, and standard output data consistent with the content attributes of the state labels; The standard dataset is divided into a training set, a validation set, and a test set according to a set ratio; the initial model is trained using the training set; the internal structure and parameters of the initial model are adjusted using the validation set; and the trained initial model is tested using the test set to obtain test metrics. Obtain the indicator threshold; compare the test indicator with the indicator threshold; if all test indicators are greater than the indicator threshold, mark the initial model as a state analysis model; otherwise, rebuild and retrain the initial model.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions for performing the steps of the method as described in any one of claims 1-9.

Images