Abnormal information monitoring method and system for battery shell stamping equipment

Abstract

The present application relates to the technical field of electric digital data processing, in particular to an abnormal information monitoring method and system of a battery shell stamping device, comprising: obtaining pressure data and vibration data in a stamping process of each time node in a set neighborhood range of the battery shell stamping device centered on a current time node, to construct a pressure data sequence and a vibration data sequence; using an improved machine learning model to output the running state of the battery shell stamping device at the next time node, and monitoring the battery shell stamping device based on the running state. The present application solves the problems of insufficient model generalization ability and high abnormal misjudgment rate.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology. More specifically, this invention relates to a method and system for monitoring abnormal information in battery casing stamping equipment. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the power battery, as one of the core components of new energy vehicles, has a direct impact on the overall vehicle quality and operational safety due to its structural strength, consistency, and safety performance. The battery casing, as a crucial load-bearing and protective component of the power battery, relies heavily on the operational stability of the stamping equipment and the controllability of the process for its forming quality. Battery casing stamping equipment typically operates continuously for extended periods under high-speed, high-frequency, and high-load conditions. The stamping process involves significant pressure changes and complex mechanical vibrations. Any abnormal equipment condition can easily lead to dimensional deviations, surface defects, or even structural hazards in the battery casing, thereby affecting subsequent assembly and the overall safety of the battery system. Therefore, accurate and timely monitoring and identification of abnormal information during the operation of battery casing stamping equipment has become a key technical issue for improving battery manufacturing quality and ensuring production safety.

[0003] In existing technologies, monitoring the operational status of stamping equipment typically relies on single sensor threshold judgments or statistical analysis methods based on empirical rules. These methods are poorly adaptable to complex operating conditions and struggle to identify potential anomalies in their early stages. With the development of machine learning, Long Short-Term Memory (LSTM) networks, due to their ability to model long-term dependencies in time series data, have been gradually introduced into the field of industrial equipment condition monitoring to learn equipment operating characteristics from time-series data such as pressure and vibration. However, existing LTM network models often employ fixed loss functions in practical applications. These loss functions typically assume that each time point contributes equally to model training, failing to fully consider the differences in physical meaning and anomaly sensitivity at different time points during the stamping process. Especially in scenarios with transient impacts, periodic fluctuations, or local anomaly propagation, fixed loss functions are easily dominated by noisy samples or data from non-critical moments, thus weakening the model's ability to learn true anomaly features.

[0004] Furthermore, the battery casing stamping process exhibits significant local temporal correlation characteristics. Anomalies often evolve gradually or erupt in a concentrated manner within a specific time neighborhood. The fixed loss function of traditional Long Short-Term Memory (LSTM) networks struggles to characterize this consistency constraint relationship based on time neighborhoods. This leads the model to tend to fit global average features during training, neglecting the dynamic changes of anomalies within local time ranges, resulting in deficiencies in the sensitivity and stability of anomaly identification. Especially when multiple data sources are involved in modeling, the fixed loss function cannot adaptively adjust its optimization direction according to the importance of pressure and vibration information under different operating conditions, easily leading to insufficient model generalization ability and a high anomaly false positive rate. Summary of the Invention

[0005] To address the problems of insufficient model generalization ability and high anomaly misclassification rate mentioned in the background art, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for monitoring abnormal information of a battery casing stamping equipment, comprising: acquiring pressure data and vibration data of the battery casing stamping equipment during the stamping process at each time node within a set neighborhood centered on the current time node, to construct a pressure data sequence and a vibration data sequence; using an improved machine learning model, outputting the operating state of the battery casing stamping equipment at the next time node, and monitoring the battery casing stamping equipment based on the operating state; wherein, the improved machine learning model includes a loss function, the loss function being positively correlated with a weight factor and the absolute value of the difference in the degree of abnormality between each time node within the set neighborhood centered on the current time node and the current time node; the weight factor of the current time node being positively correlated with the pressure data and the degree of fluctuation of the current time node, the degree of fluctuation being positively correlated with the vibration data of the current time node and negatively correlated with the standard deviation of the vibration data sequence.

[0007] The above technical solution jointly models the pressure and vibration data of the stamping process within the time neighborhood, and introduces an adaptive weight and loss constraint mechanism that is strongly correlated with the operating conditions. This makes the prediction and monitoring of the equipment's operating status more consistent with the actual physical process characteristics. The technical effect is as follows: by utilizing the temporal coupling relationship between pressure and vibration information, a forward-looking judgment of the operating status at the next moment can be achieved. At the same time, by introducing the difference in the degree of anomaly into the loss function and combining it with the weight factor that dynamically changes with the pressure intensity and vibration fluctuation, the model pays more attention to the critical moments of high risk and high fluctuation during the training and inference process. This effectively suppresses noise interference, strengthens the learning and amplification of real abnormal features, and improves the sensitivity, stability and prediction accuracy of identifying abnormal states of stamping equipment, thereby achieving reliable monitoring and early warning of the operating status of battery casing stamping equipment.

[0008] Furthermore, Loss function at time nodes for: , For Within the set neighborhood centered on the time node Weighting factors for time nodes for The degree of abnormality at each time point, For Within the set neighborhood centered on the time node The degree of abnormality at each time point, For The total number of time nodes within a defined neighborhood centered on a given time node.

[0009] The above technical solution introduces a weighted difference in the degree of anomaly of each node in the time neighborhood into the loss function, enabling the model to adaptively focus on key time nodes that are closely related to the target time and have diagnostic value during the training process. This strengthens the learning ability of abnormal evolution patterns, while suppressing the interference of instantaneous noise or non-critical fluctuations on the model, improving the sensitivity, stability and accuracy of anomaly identification, and achieving more reliable monitoring and timely early warning of the operating status of stamping equipment.

[0010] Furthermore, with Within the set neighborhood centered on the time node Weighting factors of time nodes for: , For the natural constant An exponential function with base 0. For Within the set neighborhood centered on the time node Pressure data at specific time points, To set the maximum pressure data, for The degree of fluctuation at specific points in time.

[0011] The above technical solution combines the pressure data of each node in the time neighborhood with the vibration fluctuation degree at the target time, and generates weights through an exponential function adaptive mapping. This allows data at critical moments with obvious abnormal fluctuations or high pressure to receive higher model attention, thereby effectively highlighting important information and suppressing background noise in the process of feature fusion and anomaly discrimination, and improving the sensitivity, robustness and overall monitoring accuracy of anomaly identification of stamping equipment.

[0012] Furthermore, Time nodes for: , for Vibration data at specific time points, For Within the set neighborhood centered on the time node Vibration data at specific time points, For The total number of time nodes within a defined neighborhood centered on a given time node. For The standard deviation of vibration data corresponding to all time points within a defined neighborhood centered on a given time point. These are the preset hyperparameters.

[0013] The above technical solution compares the vibration value at the target time with the average level of vibration data in its time neighborhood, and normalizes it by combining the dispersion of neighborhood vibration, thereby constructing an index that reflects the degree of local vibration deviation. This allows for effective amplification when vibration anomalies or sudden changes occur, while suppressing the interference of normal fluctuations on the judgment. As a result, it achieves sensitive capture and robust assessment of transient abnormal vibrations of stamping equipment, improving the accuracy and reliability of anomaly detection.

[0014] Furthermore, The degree of abnormality at each time point for: , For Pearson correlation coefficient function, For Within the set neighborhood centered on the time node The pressure data sequence corresponding to the time points. For Within the set neighborhood centered on the time node Vibration data sequence corresponding to the time node, For The total number of time nodes within a defined neighborhood centered on a given time node.

[0015] The above technical solution calculates the correlation between pressure and vibration sequences in the time neighborhood and uses the coupling relationship of multi-time and multi-source signals as the basis for anomaly characterization. This enables the effective capture of inconsistencies or abnormal fluctuations in pressure and vibration during the stamping process, thereby enhancing the sensitivity of anomaly identification. At the same time, by smoothing out the influence of occasional noise by comprehensively considering the average correlation value in the neighborhood, it improves the robust assessment capability of abnormal equipment operating status and the overall monitoring reliability.

[0016] Furthermore, the operating state is divided into normal state and abnormal state.

[0017] Furthermore, the pressure and vibration data are cleaned and standardized.

[0018] Furthermore, the machine learning model is a long short-term memory network model.

[0019] Furthermore, it also includes training the machine learning model, specifically: inputting the training set into the pre-built machine learning model for training; during the training process, calculating the loss between the output predicted value and the label; adjusting the model parameters using gradient descent to minimize the prediction error; iteratively adjusting the parameters of the machine learning model until the loss is less than a set value or the set number of training iterations is reached, and finally obtaining the trained machine learning model.

[0020] In a second aspect, the present invention provides an abnormal information monitoring system for a battery casing stamping equipment, comprising a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the abnormal information monitoring method for the battery casing stamping equipment described in any one of the above embodiments is implemented.

[0021] The beneficial effects of this invention are as follows:

[0022] This invention achieves dynamic monitoring and forward-looking judgment of equipment operating status by jointly collecting, cleaning, and standardizing pressure and vibration data of battery casing stamping equipment in the time neighborhood, and using a long short-term memory network model with adaptive weights for training and prediction. Its technical effect is that it can fully explore the coupling relationship between pressure and vibration signals, highlight the abnormal characteristics of key time nodes, and suppress the interference of noise and non-critical fluctuations on the model. This significantly improves the sensitivity, stability, and accuracy of abnormal identification in the stamping process, enabling timely early warning and reliable monitoring of potential equipment abnormalities. It provides effective data-driven support for ensuring the quality of battery casing production and the safe operation of equipment. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating an abnormal information monitoring method for a battery casing stamping device according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the model training MAE comparison effect of the abnormal information monitoring method for battery casing stamping equipment according to an embodiment of the present invention; Figure 3 This is a schematic block diagram illustrating the structure of an abnormal information monitoring system for a battery casing stamping equipment according to an embodiment of the present invention. Detailed Implementation

[0024] An example of a method for monitoring abnormal information in battery casing stamping equipment.

[0025] like Figure 1 As shown in the flowchart, the abnormal information monitoring method for battery casing stamping equipment according to an embodiment of the present invention includes the following steps: S1: Obtain pressure and vibration data of the battery casing stamping equipment at each time point in the stamping process within a set neighborhood centered on the current time point, in order to construct pressure data sequence and vibration data sequence.

[0026] In a preferred embodiment, the current time node of the battery casing stamping equipment is used as the time reference center. Within a pre-set time neighborhood radius, multi-time and multi-source information is collected on the stamping process to obtain stamping pressure data corresponding to each discrete time node and vibration data that is strictly synchronized with the pressure data in time, thereby constructing a pressure data sequence and a vibration data sequence with time continuity respectively. Furthermore, to address the sensor noise, transient interference, and individual abnormal sampling points that are easily introduced into stamping equipment under high-speed reciprocating motion and complex stress conditions, a systematic data cleaning process is performed on the pressure and vibration data. This includes outlier identification and removal, missing data interpolation compensation, and noise reduction based on the time or frequency domain. After data cleaning, pressure and vibration data with different physical dimensions and amplitude ranges are standardized. Dimensional differences and amplitude biases are eliminated through normalization or standard deviation standardization, ensuring that the processed pressure and vibration data sequences maintain consistency and comparability in numerical scale and statistical distribution. This effectively improves the stability and accuracy of multi-source time-series data fusion analysis, providing a reliable data foundation for subsequent stamping process status assessment, anomaly identification, and quality judgment.

[0027] S2: Using an improved machine learning model, output the operating status of the battery casing stamping equipment at the next time point.

[0028] In a preferred embodiment, the improved machine learning model includes a loss function. Loss function at time nodes for: , For Within the set neighborhood centered on the time node Weighting factors for time nodes for The degree of abnormality at each time point, For Within the set neighborhood centered on the time node The degree of abnormality at each time point, For The total number of time nodes within a defined neighborhood centered on a given time node.

[0029] By introducing a weighted loss mechanism based on temporal neighborhood consistency during model training, the core logic involves comparing the anomaly representation of the current time node with the anomaly representations of each time node in its neighborhood, and imposing differentiated penalties based on the importance of each neighborhood time node. This allows the model to focus more on key moments strongly correlated with the current stamping state during the learning process. By weighted aggregation of anomaly severity differences, the interference of abnormal fluctuations caused by instantaneous noise or occasional disturbances on model training can be suppressed. It can also strengthen the continuity and consistency constraints of real working condition changes in the time dimension, making the model more inclined to capture stable and physically meaningful anomaly evolution patterns. Thus, the constructed loss mechanism not only improves the model's sensitivity to local anomalies but also significantly enhances the robustness and generalization ability of the overall training process, which is conducive to achieving accurate characterization and reliable discrimination of abnormal states in the stamping process.

[0030] by Within the set neighborhood centered on the time node Weighting factors of time nodes for: , For the natural constant An exponential function with base 0. For Within the set neighborhood centered on the time node Pressure data at specific time points, To set the maximum pressure data, for The degree of fluctuation at specific points in time.

[0031] By coupling the pressure response at each moment within the time neighborhood with the vibration fluctuation level at the current time node, a time weighting mechanism that adaptively adjusts with changes in working conditions is constructed. The main logic is as follows: when the stamping process is running stably, the vibration fluctuation level is low, and the impact of pressure data on the overall state assessment remains relatively balanced; however, when the target moment exhibits obvious vibration fluctuation characteristics, it often means that the stamping stress state has become abnormal. At this time, by introducing the pressure response intensity and vibration fluctuation level into the exponential mapping, time nodes with higher pressure and stronger correlation with abnormal vibration are given higher weights, thereby highlighting key stamping behaviors and weakening background or redundant information within the time neighborhood. This nonlinear weighting construction method not only enhances the sensitivity to abnormal stamping events but also effectively avoids the problem of insufficient discrimination of linear weighting under extreme working conditions, making subsequent time series feature fusion and anomaly discrimination more focused on key moments with diagnostic value, thus improving the accuracy and engineering robustness of the overall detection method.

[0032] Time nodes for: , for Vibration data at specific time points, For Within the set neighborhood centered on the time node Vibration data at specific time points, For The total number of time nodes within a defined neighborhood centered on a given time node. For The standard deviation of vibration data corresponding to all time points within a defined neighborhood centered on a given time point. These are the preset hyperparameters.

[0033] Taking the vibration response at the current time point as the core, this paper constructs a quantitative index reflecting the degree of local deviation by comparing the vibration value at that moment with the overall statistical characteristics of vibration data in its time neighborhood. The main logic is that under normal stamping conditions, the vibration signal usually exhibits a relatively stable statistical distribution in a short time neighborhood. However, when the equipment status or stamping process is abnormal, the vibration response at the target time often deviates significantly from the average level in its neighborhood. Therefore, by using the average characteristics of the neighborhood vibration data as a benchmark and combining it with the normalization processing of the neighborhood vibration dispersion, it can not only accurately characterize the degree of abrupt change in the vibration behavior at the target time relative to the local background, but also effectively suppress the scale effect caused by the change in vibration amplitude under different working conditions. The introduction of preset parameters further avoids the problem of index instability in scenarios with small vibration fluctuations, thereby achieving sensitive detection and robust evaluation of transient abnormal vibrations and improving the reliability and engineering adaptability of abnormal identification in the stamping process.

[0034] The degree of abnormality at each time point for: , For Pearson correlation coefficient function, For Within the set neighborhood centered on the time node The pressure data sequence corresponding to the time points. For Within the set neighborhood centered on the time node Vibration data sequence corresponding to the time node, For The total number of time nodes within a defined neighborhood centered on a given time node.

[0035] By comprehensively characterizing the correlation between pressure and vibration signals during the stamping process within the time neighborhood of the current time point, and using the coupling degree between multiple physical quantities at multiple times and sources as the criterion for anomaly detection, the main logic is as follows: Under normal stamping conditions, pressure changes and vibration responses usually have relatively stable and consistent correlation characteristics. However, when material abnormalities, die wear, or force imbalances occur during the stamping process, the intrinsic correlation between the two types of signals will be significantly disturbed. Therefore, by comprehensively calculating the correlation between the pressure data sequence and the corresponding vibration data sequence at each time point within the time neighborhood, and using the average result to characterize the degree of anomaly at the current time point, the influence of instantaneous noise and occasional fluctuations can be effectively reduced, making the anomaly assessment results smoother and more robust. This enables sensitive perception and accurate characterization of potential abnormal states in the stamping process, improving the reliability of anomaly identification and its engineering practical value.

[0036] The machine learning model is a long short-term memory network model.

[0037] Furthermore, the machine learning model is trained. For example, a Long Short-Term Memory (LSTM) network model is trained as follows: First, a training set is constructed and obtained. This training set includes the anomaly levels and weight factors of each time point within a defined neighborhood centered on the current time point, with labels manually assigned. The training set is then input into the pre-constructed LTM network model for training. During training, the loss between the output predicted value and the label is calculated; the model parameters are adjusted using gradient descent to minimize the prediction error; the parameters of the LTM network model are iteratively adjusted until the loss is less than a set value or a set number of training iterations are reached, ultimately resulting in a trained LTM network model.

[0038] like Figure 2 The figure shows a comparison of the Model-Based Equation (MAE) training results of the abnormal information monitoring method for battery casing stamping equipment according to an embodiment of the present invention.

[0039] S3: Monitor the battery casing stamping equipment based on the operating status.

[0040] In a preferred embodiment, the operating state is divided into a normal state and an abnormal state. Based on the obtained operating state evaluation results, the working process of the battery casing stamping equipment is continuously monitored and dynamically judged. The operating state is used to characterize the overall working health level of the equipment under the current working conditions and serves as an important basis for the safe operation and quality control of the equipment. In a preferred embodiment, by comprehensively analyzing multi-source time-series data and anomaly characterization results during the stamping process, the equipment operating status is divided into normal and abnormal states. When the monitoring results indicate that the equipment is in a normal state, it means that there is a stable and coordinated relationship between the stamping pressure and vibration response, the equipment operates smoothly, and the stamping quality is controllable. When the monitoring results show an abnormal state, it indicates that there is an imbalance in the pressure-vibration coupling relationship, increased fluctuations, or an abnormal evolution trend during the stamping process, which may correspond to potential risks such as mold wear, material abnormalities, or mechanical deviations. Through continuous monitoring and state classification of the operating status, not only can real-time perception and early warning of the operating conditions of the battery casing stamping equipment be achieved, but also a reliable basis can be provided for subsequent equipment maintenance decisions, process parameter adjustments, and production quality assurance, thereby effectively improving the safety, stability, and overall production efficiency of the stamping equipment.

[0041] The present invention performs multi-source time-series modeling of pressure and vibration data of battery casing stamping equipment in the time neighborhood, and combines adaptive weights and anomaly degree-driven loss functions to enable the machine learning model to dynamically focus on abnormal features at key moments and suppress noise interference. At the same time, it uses a long short-term memory network to capture the time dependence in the stamping process, and achieves accurate prediction and classification of the operating state at the next moment through training and optimization. Thus, in equipment operation monitoring, it can perceive the evolution of anomalies in real time, accurately identify potential risks, and provide a reliable basis for equipment maintenance, process adjustment and production quality control, effectively improving the safety, stability and overall production efficiency of the stamping process.

[0042] Example of an abnormal information monitoring system for battery casing stamping equipment: like Figure 3 As shown in the figure, the structural block diagram of the abnormal information monitoring system of the battery casing stamping equipment according to an embodiment of the present invention includes a processor and a memory.

[0043] This invention also provides an abnormal information monitoring system for battery casing stamping equipment. For example... Figure 3 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the abnormal information monitoring method for the battery casing stamping equipment according to the present invention.

[0044] The abnormal information monitoring system of the battery casing stamping equipment also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0045] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0046] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0047] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for monitoring abnormal information in battery casing stamping equipment, characterized in that, include: The pressure and vibration data of the battery casing stamping equipment during the stamping process are acquired within a set neighborhood centered on the current time node to construct a pressure data sequence and a vibration data sequence. Using an improved machine learning model, the operating status of the battery casing stamping equipment at the next time node is output, and the battery casing stamping equipment is monitored based on the operating status. The improved machine learning model includes a loss function that is positively correlated with the weight factor and the absolute value of the difference in the degree of abnormality between each time node in a set neighborhood centered on the current time node and the current time node. The weighting factor at the current time point is positively correlated with the pressure data and fluctuation level at the current time point. The fluctuation level is positively correlated with the vibration data at the current time point and negatively correlated with the standard deviation of the vibration data sequence.

2. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, Loss function at time nodes for: , For Within the set neighborhood centered on the time node Weighting factors for time nodes for The degree of abnormality at each time point, For Within the set neighborhood centered on the time node The degree of abnormality at each time point, For The total number of time nodes within a defined neighborhood centered on a given time node.

3. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, by Within the set neighborhood centered on the time node Weighting factors of time nodes for: , For the natural constant An exponential function with base 0. For Within the set neighborhood centered on the time node Pressure data at specific time points, To set the maximum pressure data, for The degree of fluctuation at specific points in time.

4. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, Time nodes for: , for Vibration data at specific time points, For Within the set neighborhood centered on the time node Vibration data at specific time points, For The total number of time nodes within a defined neighborhood centered on a given time node. For The standard deviation of vibration data corresponding to all time points within a defined neighborhood centered on a given time point. These are the preset hyperparameters.

5. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, The degree of abnormality at each time point for: , For Pearson correlation coefficient function, For Within the set neighborhood centered on the time node The pressure data sequence corresponding to the time points. For Within the set neighborhood centered on the time node Vibration data sequence corresponding to the time node, For The total number of time nodes within a defined neighborhood centered on a given time node.

6. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, The operating status is divided into normal status and abnormal status.

7. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, The pressure and vibration data are cleaned and standardized.

8. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, The machine learning model is a long short-term memory network model.

9. The method for monitoring abnormal information of battery casing stamping equipment according to claim 1, characterized in that, This also includes training machine learning models, specifically: The training set is fed into a pre-built machine learning model for training. During the training process, the loss between the output prediction value and the label is calculated. Use gradient descent to adjust model parameters to minimize prediction error; The parameters of the machine learning model are iteratively adjusted until the loss is less than a set value or the set number of training iterations are reached, and finally a well-trained machine learning model is obtained.

10. An abnormal information monitoring system for battery casing stamping equipment, characterized in that, The device includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the abnormal information monitoring method for the battery casing stamping equipment according to any one of claims 1 to 9.

Images