CN-122017574-A - Method and system for ultrasonically detecting state of charge of hard-shell lithium battery based on deep learning

CN122017574ACN 122017574 ACN122017574 ACN 122017574ACN-122017574-A

Abstract

The invention discloses a method and a system for detecting the state of charge of a hard-shell lithium battery by ultrasonic based on deep learning, and relates to the field of industrial detection, wherein the method comprises the steps of obtaining an ultrasonic time domain signal of the hard-shell lithium battery, and preprocessing the ultrasonic time domain signal to obtain a preprocessed ultrasonic time domain signal; the method comprises the steps of constructing an initial battery state of charge estimation model based on a multi-scale convolutional neural network technology, a two-way long-short-term memory network algorithm and a selective state space mechanism, inputting a preprocessed ultrasonic time domain signal into the initial battery state of charge estimation model to perform training calculation to obtain training iteration times and a loss error threshold value, and performing iterative calculation on the initial battery state of charge estimation model by using the training iteration times and the loss error threshold value to obtain an optimal battery state of charge estimation model so as to improve estimation accuracy of the lithium battery state of charge. According to the invention, the ultrasonic transmission signal is used as input data, and the estimation accuracy of the lithium battery state of charge is improved by utilizing the information contained in the signal.

Inventors

ZHENG CHICHAO
CHU WENWEN
ZHANG YUSHAN
XIA QIUSHI
WANG HAORAN

Assignees

合肥工业大学
恒钧检测技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (10)

1. The method for ultrasonically detecting the charge state of the hard-shell lithium battery based on deep learning is characterized by comprising the following steps of: s1, acquiring an ultrasonic time domain signal of a hard shell lithium battery, and preprocessing the ultrasonic time domain signal to obtain an ultrasonic time domain signal; S2, constructing an initial battery state-of-charge estimation model based on a multi-scale convolutional neural network technology, a two-way long-short-term memory network algorithm and a selective state space mechanism; S3, inputting the preprocessed ultrasonic time domain signals into an initial battery state of charge estimation model for training calculation to obtain training iteration times and a loss error threshold; And S4, carrying out iterative computation on the initial battery state-of-charge estimation model by using the training iteration times and the loss error threshold value to obtain an optimal battery state-of-charge estimation model so as to improve the estimation accuracy of the lithium battery state-of-charge.
2. The method for detecting the state of charge of the hard-shell lithium battery based on deep learning according to claim 1, wherein the steps of obtaining the ultrasonic time-domain signal of the hard-shell lithium battery, and preprocessing the ultrasonic time-domain signal to obtain the preprocessed ultrasonic time-domain signal comprise: S11, acquiring ultrasonic transmission time domain signals of the hard shell lithium battery under different percentage electric quantities; s12, processing and analyzing the ultrasonic transmission signals, and determining an optimal signal of the middle measuring point according to an experimental result; s13, positioning the selected normalized experimental signal to obtain the maximum value of the experimental signal, and determining the coordinate point of the first zero crossing point after the maximum value to serve as the coordinate point of signal extraction; S14, dividing an ultrasonic transmission signal into a preamble signal and a follow-up signal based on coordinate points extracted by the signal, wherein the follow-up signal consists of data after the coordinates are extracted and data after the preset times of zero crossing points and before the last zero crossing point are continuously met; S15, carrying out batch processing on the extracted subsequent signal data, and taking preset data as a batch to serve as the preprocessed ultrasonic time domain signal.
3. The deep learning based ultrasonic hard-shell lithium battery state of charge detection method of claim 1, wherein the multi-scale convolutional neural network technique comprises: The multi-scale convolutional neural network is composed of two convolutional neural networks with different preset convolutional kernel sizes; Respectively extracting the preprocessed ultrasonic time domain signals by using two convolution neural networks with different preset convolution kernel sizes to obtain a first frequency characteristic and a second frequency characteristic; Each convolutional neural network is formed by two convolutional layers and two average pooling layers alternately, wherein the average pooling layers set the pooling kernel size as a preset value according to the channel number of an input signal, ensure that the characteristics of the input signal can participate in calculation, and set the pooling step length as the preset value; and carrying out nonlinear transformation on the extracted local features by using a nonlinear activation function to obtain the spatial features of the ultrasonic time domain signals.
4. The deep learning-based ultrasonic hard-shell lithium battery state-of-charge detection method of claim 3, wherein the two-way long-short-term memory network algorithm comprises: the two-way long-short-term memory network algorithm consists of a forward long-short-term memory network and a backward long-short-term memory network; the pre-processed ultrasonic time domain signals are processed by the forward long-short-term memory network according to time sequence, so that past context information of the sequence is captured, and a forward hidden state is output; The backward long-short-term memory network processes the preprocessed ultrasonic time domain signals in time reverse order to capture future context information of the sequence and output a backward hidden state; and fusing the forward hiding state and the backward hiding state at a time point to obtain the time characteristics of the preprocessed ultrasonic time domain signal.
5. The deep learning based ultrasound detection hard shell lithium battery state of charge method of claim 4, wherein the selective state space mechanism comprises: extracting and fusing spatial features and temporal features extracted by a multi-scale convolutional neural network technology and a two-way long-short-term memory network algorithm by utilizing a selective state space mechanism; calculating the preprocessed ultrasonic time domain signals through a selective scanning mode and a dynamic parameter matrix in a selective state space mechanism; and processing the calculation result by utilizing a selective dynamic weight adjustment mode and state evolution integration to obtain the optimized characteristics of the preprocessed ultrasonic time domain signal.
6. The deep learning-based ultrasonic hard-shell lithium battery state-of-charge detection method of claim 5, wherein constructing an initial battery state-of-charge estimation model based on a multi-scale convolutional neural network technique, a two-way long-short-term memory network algorithm and a selective state-space mechanism comprises: s21, constructing an initial battery state-of-charge estimation model according to a multi-scale convolutional neural network technology, a two-way long-short-term memory network algorithm and a selective state space mechanism; s22, extracting spatial features of the preprocessed ultrasonic time domain signals by using a multi-scale convolutional neural network technology; S23, extracting the time characteristics of the preprocessed ultrasonic time domain signals by using a two-way long-short-term memory network algorithm; s24, extracting optimized features of the preprocessed ultrasonic time domain signals by using a selective state space mechanism.
7. The deep learning based ultrasound detection hard shell lithium battery state of charge method of claim 6, wherein the initial battery state of charge estimation model further comprises: an attention classification output mechanism; Calculating the similarity of each element in the preprocessed ultrasonic time domain signal with other elements by using a self-attention mechanism so as to capture the dependency relationship among all the positions in the sequence; Generating a query, a key and a value matrix by using a self-attention mechanism, calculating attention scores by dot products, and carrying out weighted summation to obtain new data representation of each element; And performing characteristic transformation and mapping on the processed ultrasonic time domain signal by using linear transformation and nonlinear activation functions to obtain an estimation result of the charge state of the hard-shell lithium battery so as to realize classification output.
8. The method for ultrasonically detecting the state of charge of the hard-shell lithium battery based on deep learning according to claim 1, wherein the step of inputting the preprocessed ultrasonic time domain signal into the initial battery state of charge estimation model for training calculation to obtain the training iteration times and the loss error threshold value comprises the following steps: S31, constructing a loss function by using the first moment estimation after deviation correction, the second moment estimation after deviation correction and the learning rate; S32, inputting the value of the first moment estimation after the deviation correction into the loss function calculation again to obtain the second moment estimation after the deviation correction; And S33, training the initial battery state of charge estimation model based on the loss function, and reaching the preset training iteration times and the loss error threshold.
9. The method for ultrasonically detecting a hard-shell lithium battery state of charge based on deep learning according to claim 1, wherein the iterative calculation of the initial battery state of charge estimation model by using the training iteration number and the loss error threshold value to obtain an optimal battery state of charge estimation model, so as to improve the estimation precision of the lithium battery state of charge comprises: s41, dividing the data set into a training set, a verification set and a test set according to a preset proportion; S42, presetting a learning rate of the self-adaptive optimizer as an initial value, setting a weight attenuation value, adding L2 regularization in the self-adaptive optimizer for preventing over-fitting, and setting a regularization rate; S43, removing the maximum value and the minimum value in the cross verification based on the preset value of the cross verification, and calculating the average value of the residual result; s44, performing iterative training on the initial battery state-of-charge estimation model based on a training strategy and a data set, and calculating by using a loss function; And S45, when the training iteration times reach the preset times and the loss error is smaller than a preset threshold value, training is stopped, and an optimal battery state-of-charge estimation model is obtained so as to improve the estimation accuracy of the lithium battery state-of-charge.
10. A deep learning based ultrasonic hard-shell lithium battery state of charge detection system for implementing the deep learning based ultrasonic hard-shell lithium battery state of charge detection method of any one of claims 1-9, comprising: The pretreatment module is used for acquiring an ultrasonic time domain signal of the hard shell lithium battery, and carrying out pretreatment on the ultrasonic time domain signal to obtain a pretreated ultrasonic time domain signal; the model construction module is used for constructing an initial battery state-of-charge estimation model based on a multi-scale convolutional neural network technology, a two-way long-short-term memory network algorithm and a selective state space mechanism; the model training module is used for inputting the preprocessed ultrasonic time domain signals into the initial battery state of charge estimation model to carry out training calculation, so as to obtain training iteration times and a loss error threshold; And the model optimization module is used for carrying out iterative computation on the initial battery state of charge estimation model by utilizing the training iteration times and the loss error threshold value to obtain an optimal battery state of charge estimation model so as to improve the estimation precision of the lithium battery state of charge.

Description

Method and system for ultrasonically detecting state of charge of hard-shell lithium battery based on deep learning Technical Field The invention relates to the field of industrial detection, in particular to a method and a system for ultrasonically detecting the state of charge of a hard-shell lithium battery based on deep learning. Background With the rapid development of new energy products, lithium batteries have become an indispensable mainstream energy storage technology, which makes it important to accurately estimate the state of charge of lithium ion batteries. The state of charge (SoC, stateofCharge) of a lithium battery, also called the residual capacity, refers to the ratio S of the residual capacity of the lithium battery after being used for a period of time or being left unused for a long time to the rated capacity under the same conditions, ultrasonic battery detection is a recognized method for obtaining feature-rich data from the battery, and the ultrasonic technology is non-invasive, non-destructive and convenient. In addition, the ultrasonic test is different from the test method based on the electrical parameters, no extra circuit is needed, and the complexity of the battery test is simplified. Increasingly researchers are using ultrasound technology to estimate the SOC of a battery. The current technology for detecting the battery by utilizing the ultrasonic wave comprises the ultrasonic transmission technology, the ultrasonic reflection technology and the ultrasonic guided wave technology. Because of the complex internal structure and physical characteristics of the battery, the ultrasonic reflection signals and the guided wave signals are complex, the characteristics of the reflection signals and the guided wave signals are complex and various under the condition that defects exist in the battery, but compared with other methods, the ultrasonic transmission method does not need to consider the internal structure of the battery, and the characteristics of the signal waveform are simple. The existing measurement method based on ultrasonic transmission is to detect a soft package lithium battery with low capacity. The battery core of the soft-packed battery is thin, the thickness of the shell is also thin, so that ultrasonic transmission signals related to the electric quantity information in the battery are easy to obtain, the battery core of the hard-shell battery with large capacity generally adopts a winding structure, and a steel shell with certain thickness is arranged outside the battery core, so that the attenuation of ultrasonic waves in the thicker hard-shell lithium battery is serious, the signal-to-noise ratio of signals acquired by a receiving transducer is low, and the traditional signal processing method is difficult to accurately reflect the SOC of the lithium battery. Patent CN202311018433.5 provides a device and a method for nondestructive testing of lithium batteries based on ultrasonic resonance spectrum, which are characterized by precisely measuring ultrasonic resonance spectrums based on transmitted waves or reflected waves at different positions of the lithium batteries, identifying a plurality of resonance frequencies from the ultrasonic resonance spectrums and analyzing distribution characteristics of the ultrasonic resonance spectrums, analyzing characteristics of pass bands and forbidden bands in ultrasonic resonance spectrum signals, extracting amplitude information of the transmitted waves or the reflected waves at different frequency points from the ultrasonic resonance spectrum signals, and obtaining amplitude cloud charts of the transmitted waves or the reflected waves of the lithium batteries at the plurality of frequency points and mechanical quality factors of the lithium batteries. In particular, compared with SoC detection by utilizing the maximum value change of the amplitude of the pulse ultrasonic transmission wave time domain waveform, the method can accurately obtain amplitude attenuation of a plurality of single-frequency ultrasonic waves after penetrating through a lithium battery, and can extract a mechanical quality factor value from a resonance peak. The propagation attenuation of different frequency waves in the same medium is inconsistent, the mechanical quality factor value directly reflects the damping of the material, the amplitude attenuation information of a plurality of frequency ultrasonic waves penetrating through the lithium battery is synthesized, and the mechanical quality factor value is combined to quantitatively evaluate the SoC. The method essentially utilizes amplitude information of different frequency points to reflect the SOC, and the mode does not consider the whole time domain and frequency domain information of the transmission signal and the correlation between the front waveform and the rear waveform of the transmission signal. The estimation result of SOC is not disclosed in this patent. When the lithium ion batt