A system and method for monitoring the operating status of an automatic cell flipping device.

Abstract

This application provides a system and method for monitoring the operating status of an automatic cell flipping device, relating to the field of intelligent monitoring technology for industrial equipment. The system decouples the time-frequency domain features of the multi-source heterogeneous monitoring dataset of the automatic cell flipping device to obtain an operating status feature vector. Based on this feature vector, a health baseline model for the equipment is established. The parameters of the health baseline model are adaptively tuned to obtain state deviation residuals. Multi-scale confidence fusion is performed on the state deviation residuals based on real-time monitoring data to generate a comprehensive health index sequence. An anomaly level classification is performed on the comprehensive health index sequence based on an entropy weight decision mechanism to obtain an anomaly level classification result. Based on the anomaly level classification result, maintenance instructions are driven to achieve pre-maintenance scheduling and fault location for the automatic cell flipping device. This application can achieve accurate assessment of the operating status of the automatic cell flipping device and early warning and precise fault location, thereby improving the maintenance efficiency of battery production equipment.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring technology for industrial equipment, and more specifically, to a system and method for monitoring the operating status of an automatic cell flipping device. Background Technology

[0002] With the continuous improvement of industrial intelligence, intelligent monitoring of industrial equipment has become a core technical means to ensure production safety, improve operating efficiency, and achieve predictive maintenance. By collecting and analyzing multi-source signals (such as vibration, current, temperature, images, etc.) during equipment operation, its health status can be assessed in real time and potential faults can be warned.

[0003] The automatic cell flipping device is a core piece of equipment in key processes such as electrode stacking and cell assembly. Its operating status directly affects the alignment accuracy of the cells, production cycle time, and final product quality. Therefore, real-time and accurate status monitoring and fault diagnosis of the automatic cell flipping device are crucial for ensuring the continuous, stable, and efficient operation of battery production lines. However, current mainstream technologies have the following limitations: First, the monitoring methods are relatively simple, usually relying on single vibration alarms or current over-limit judgments, failing to effectively integrate multi-dimensional information such as visual data, resulting in insensitivity to complex faults or early subtle anomalies. Second, the status assessment models often rely on fixed thresholds or static empirical models, making it difficult to adapt to changes caused by equipment wear and operating condition fluctuations, leading to high false alarm and false negative rates. Furthermore, there is a lack of in-depth data mining and intelligent decision-making based on monitoring data, failing to extract multi-scale features from time-series signals, and failing to achieve decision-making upgrades from abnormal alarms to level classification and root cause inference. Therefore, how to achieve accurate assessment of the operating status of the automatic cell flipping device and early warning and precise fault location to improve the operation and maintenance efficiency of battery production equipment is a challenge facing the industry. Summary of the Invention

[0004] This application provides a system and method for monitoring the operating status of an automatic cell flipping device, which can achieve accurate assessment of the operating status of the automatic cell flipping device, early warning of faults, and precise location, thereby improving the operation and maintenance efficiency of battery production equipment.

[0005] In a first aspect, this application provides a method for monitoring the operating status of an automatic cell flipping device, the monitoring method comprising the following steps: The operating parameters and image data of the automatic cell flipping device are collected to obtain a multi-source heterogeneous monitoring dataset; The time-frequency domain features of the multi-source heterogeneous monitoring dataset are decoupled to obtain the operating status feature vector, and a device health baseline model is established based on the operating status feature vector. The parameters of the device health baseline model are adaptively tuned using a genetic optimization strategy to obtain the state deviation residuals. Real-time monitoring data of the automatic cell flipping device is obtained. Based on the real-time monitoring data, multi-scale confidence fusion is performed on the state deviation residual to generate a comprehensive health index sequence. Based on the entropy weight decision mechanism, the comprehensive health index sequence is classified into abnormal levels to obtain the abnormal level classification result. Based on the abnormality level classification results, the operation and maintenance instructions are driven to realize the pre-maintenance scheduling and fault location of the automatic cell flipping device.

[0006] In this embodiment, the collection of operating parameters and image data from the automatic cell flipping device to obtain a multi-source heterogeneous monitoring dataset specifically includes: The three-phase current signal and vibration acceleration signal of the automatic cell flipping device are collected, and the time-series image frames during the movement of the flipping robotic arm are captured. The three-phase current signal is subjected to Clarke transform to obtain the current component in the two-phase stationary coordinate system. The vibration acceleration signal, the current component and the time sequence image frame are used to form the original dataset. The original dataset is subjected to multi-channel synchronous calibration and timestamp alignment to generate a multi-source heterogeneous monitoring dataset.

[0007] In this embodiment, the time-frequency domain feature decoupling of the multi-source heterogeneous monitoring dataset to obtain the running state feature vector specifically includes: Variational mode decomposition is performed on the vibration signal components in the multi-source heterogeneous monitoring dataset, and then the frequency domain energy distribution characteristics are obtained by Hilbert transform. Optical flow field calculations are performed on the image frame sequences in the multi-source heterogeneous monitoring dataset to extract statistical features of the motion vector field; The Park transform is performed on the current components in the multi-source heterogeneous monitoring dataset to obtain the electrical state characteristics; The frequency domain energy distribution characteristics, the statistical characteristics of the motion vector field, and the electrical state characteristics are fused by principal component analysis to obtain the operating state feature vector.

[0008] In this embodiment, establishing a device health baseline model based on the operating state feature vector specifically includes: A normal operating condition sample set is constructed based on the aforementioned operating state feature vectors; Using a support vector machine with the normal operating condition sample set as training data, find the smallest hypersphere; A health baseline model for the device is constructed based on the center and radius of the minimum hypersphere.

[0009] In this embodiment, the parameters of the device health baseline model are adaptively tuned using a genetic optimization strategy to obtain the state deviation residuals, specifically including: The fitness function is determined based on the device health baseline model. When the fitness function converges, the device health baseline model is retrained using the genetic individuals extracted from the support vector machine, thereby obtaining the state update vector; The state deviation residual is determined by the degree of deviation between the state update vector and the center of the smallest hypersphere; The genetic individuals extracted by the support vector machine include the kernel width parameter and the error penalty factor.

[0010] In this embodiment, real-time monitoring data of the automatic cell flipping device is obtained through vibration sensors, current sensors, and industrial cameras.

[0011] In this embodiment, the process of generating a comprehensive health index sequence by performing multi-scale confidence fusion on the state deviation residuals based on the real-time monitoring data specifically includes: The state deviation residual is subjected to dual-tree complex wavelet transform to obtain sub-band coefficients at different scales; Calculate the energy entropy of the subband coefficients at each scale to obtain the local confidence at different scales; The weights of each scale are dynamically allocated based on the local confidence level at different scales, and then a comprehensive health index sequence is calculated based on all weights.

[0012] In this embodiment, the anomaly level classification of the comprehensive health index sequence based on the entropy weight decision mechanism specifically includes: Determine the sample entropy and root mean square of the comprehensive health index sequence; The objective weights of the sample entropy and the root mean square are calculated using the entropy weight method. Construct a membership function based on the sample entropy and the objective weights of the root mean square; The anomaly level classification result is determined by the membership function and the preset anomaly level.

[0013] In this embodiment, the pre-maintenance scheduling and fault location of the automatic cell flipping device are specifically implemented by driving maintenance instructions based on the anomaly level classification results: Based on the anomaly level classification results, operation and maintenance instructions are generated, and key scale features and sensor channel information that triggered the anomaly level classification results are traced back synchronously. By combining the key scale features and the sensor channel information, the faulty components and causes are queried to obtain fault location information; The operation and maintenance instructions enable pre-maintenance scheduling of the automatic cell flipping device, and the fault location information enables fault location of the automatic cell flipping device.

[0014] Secondly, this application provides a battery cell automatic flipping device operation status monitoring system for executing a battery cell automatic flipping device operation status monitoring method, the status monitoring system comprising: The data synchronization module is used to collect the operating parameters and image data of the automatic cell flipping device, thereby obtaining a multi-source heterogeneous monitoring dataset. The feature decoupling module is used to decouple the time-frequency domain features of the multi-source heterogeneous monitoring dataset to obtain the operating state feature vector, and to establish a device health baseline model based on the operating state feature vector. The residual generation module is used to adaptively tune the parameters of the device health baseline model through a genetic optimization strategy to obtain the state deviation residual. The multi-scale entropy weight decision module is used to acquire real-time monitoring data of the automatic cell reversal device, perform multi-scale confidence fusion on the state deviation residual based on the real-time monitoring data, generate a comprehensive health index sequence, and classify the comprehensive health index sequence into anomaly levels based on the entropy weight decision mechanism to obtain the anomaly level classification result. The operation and maintenance scheduling module is used to drive operation and maintenance instructions based on the anomaly level classification results to realize the pre-maintenance scheduling and fault location of the automatic cell flipping device.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The system collects operating parameters and image data of the automatic cell flipping device to obtain a multi-source heterogeneous monitoring dataset. It then decouples the time-frequency domain features of the dataset to obtain an operating state feature vector, and establishes a device health baseline model based on this vector. The parameters of the health baseline model are adaptively tuned using a genetic optimization strategy to obtain state deviation residuals. Real-time monitoring data of the automatic cell flipping device is acquired, and multi-scale confidence fusion is performed on the state deviation residuals based on the real-time monitoring data to generate a comprehensive health index sequence. An anomaly level classification is then performed on the comprehensive health index sequence based on an entropy weight decision mechanism to obtain an anomaly level classification result. Finally, maintenance instructions are driven based on the anomaly level classification result to achieve pre-maintenance scheduling and fault location for the automatic cell flipping device.

[0016] Therefore, this application can achieve accurate assessment of the operating status of the automatic cell reversal device and early warning and precise location of faults. First, by synchronously collecting multi-source heterogeneous monitoring datasets and performing multi-channel synchronous calibration and timestamp alignment, a comprehensive, synchronous, and highly consistent data foundation is built for status assessment. Furthermore, by decoupling the time-frequency domain features of the aforementioned datasets and fusing them through principal component analysis, operating status feature vectors that can deeply characterize the comprehensive mechanical, visual, and electrical status of the equipment are extracted. Then, a health baseline model of the equipment is constructed based on a support vector machine, achieving a high-dimensional, accurate mathematical description of complex operating states and a quantitative definition of normal boundaries. Second, by adaptively tuning the baseline model parameters through a genetic optimization strategy, it can dynamically adapt to the slow changes in the equipment itself. By analyzing the fluctuations in operating conditions and utilizing the residuals of state deviations, the sensitivity of early fault detection can be improved. Then, a dual-tree complex wavelet transform is performed on the residuals of state deviations to achieve multi-scale decomposition. Local confidence is formed by calculating the energy entropy and peak factor at each scale, and a comprehensive health index sequence is generated by dynamically weighting and fusing these values. Finally, based on the entropy weight decision mechanism, the sample entropy and root mean square of the index sequence are objectively weighted and their membership is determined. This facilitates the leap from single-point residuals to multi-scale reliable assessment and then to data-driven, objectively graded intelligent decision-making. Finally, the operation and maintenance instructions are driven based on the anomaly level classification results, and key scale features and sensor channel information are backtracked simultaneously. The intelligent analysis results are directly and accurately converted into graded pre-maintenance scheduling actions and specific fault troubleshooting guidelines.

[0017] In summary, the technical solution adopted in this application can achieve accurate assessment of the operating status of the automatic cell flipping device, early warning of faults, and precise location, thereby improving the operation and maintenance efficiency of battery production equipment. 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 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 flowchart of a method for monitoring the operating status of an automatic cell flipping device according to this application; Figure 2 This is an exemplary flowchart for determining the feature vector of the running state according to the present application; Figure 3 This is an exemplary flowchart for determining a comprehensive health index sequence according to the present application; Figure 4This is a module structure diagram of an automatic cell flipping device operation status monitoring system provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of this application.

[0021] This application provides a system and method for monitoring the operational status of an automatic cell flipping device. The core of this system involves collecting operational parameters and image data from the automatic cell flipping device to obtain a multi-source heterogeneous monitoring dataset. The multi-source heterogeneous monitoring dataset is then decoupled from its time-frequency domain features to obtain an operational status feature vector. Based on this feature vector, a device health baseline model is established. The parameters of the health baseline model are adaptively tuned using a genetic optimization strategy to obtain state deviation residuals. Real-time monitoring data of the automatic cell flipping device is acquired. Based on this real-time monitoring data, multi-scale confidence fusion is performed on the state deviation residuals to generate a comprehensive health index sequence. An entropy weight decision mechanism is used to classify the comprehensive health index sequence into anomaly levels, resulting in anomaly level classification results. Based on these anomaly level classification results, maintenance instructions are driven to achieve pre-maintenance scheduling and fault location for the automatic cell flipping device.

[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is a flowchart of a method for monitoring the operating status of an automatic cell flipping device according to this embodiment of the present application. The monitoring method includes the following steps: In step S1, the operating parameters and image data of the automatic cell flipping device are collected to obtain a multi-source heterogeneous monitoring dataset.

[0023] In this embodiment, the operation parameters and image data of the automatic cell flipping device are collected to obtain a multi-source heterogeneous monitoring dataset, which can be achieved in the following manner: The three-phase current signal and vibration acceleration signal of the automatic cell flipping device are collected, and the time-series image frames during the movement of the flipping robotic arm are captured. The three-phase current signal is subjected to Clarke transform to obtain the current component in the two-phase stationary coordinate system. The vibration acceleration signal, the current component and the time sequence image frame are used to form the original dataset. The original dataset is subjected to multi-channel synchronous calibration and timestamp alignment to generate a multi-source heterogeneous monitoring dataset.

[0024] In practice, firstly, the three-phase current signals of the automatic cell flipping device are acquired using a three-phase current sensor, the vibration acceleration signals of the automatic cell flipping device are acquired using a vibration acceleration sensor, and time-series image frames of the automatic cell flipping device are captured using an industrial area array camera. Then, the instantaneous values ​​of the three-phase currents are substituted into the Clarke transform matrix formula for calculation, and the calculation results are used as current components. The vibration acceleration signals, current components, and time-series image frames are then aggregated according to the same time sequence to form the original dataset. Finally, for channels with different sampling frequencies (such as current signals and time-series image frames), a linear interpolation algorithm is used to resample the data onto a unified high-frequency time grid. The dataset after the above time alignment processing is used as a multi-source heterogeneous monitoring dataset.

[0025] It should be noted that the multi-source heterogeneous monitoring dataset in this application refers to a standardized data set of signals from three physical sources: mechanical vibration, motor electrical signals, and machine vision signals. Among them, the vibration acceleration signal represents the dynamic force and state of the mechanical structure; the current component obtained by Clarke transform is used to analyze the load characteristics and energy conversion efficiency of the motor; and the time-series image frames provide the motion process and position information of the automatic cell flipping device.

[0026] In step S2, the time-frequency domain features of the multi-source heterogeneous monitoring dataset are decoupled to obtain the operating status feature vector, and a device health baseline model is established based on the operating status feature vector.

[0027] Preferably, in this embodiment, reference Figure 2 As shown, this diagram is an exemplary flowchart for determining the operating state feature vector according to the present application. In this embodiment, the time-frequency domain feature decoupling of the multi-source heterogeneous monitoring dataset to obtain the operating state feature vector can be achieved through the following steps: First, in step S21, variational mode decomposition is performed on the vibration signal components in the multi-source heterogeneous monitoring dataset, and then the frequency domain energy distribution characteristics are obtained by Hilbert transform. Secondly, in step S22, optical flow field calculation is performed on the image frame sequence in the multi-source heterogeneous monitoring dataset to extract the statistical features of the motion vector field; Then, in step S23, Park transformation is performed on the current components in the multi-source heterogeneous monitoring dataset to obtain electrical state characteristics; Finally, in step S24, the frequency domain energy distribution characteristics, the statistical characteristics of the motion vector field, and the electrical state characteristics are fused by principal component analysis to obtain the operating state feature vector.

[0028] In practical implementation, firstly, variational mode decomposition (VMD) is an algorithm that adaptively decomposes a complex signal into several narrowband components with different center frequencies. Iteratively searching to minimize the bandwidth of each component yields a set of mutually distinguishable intrinsic mode functions (EMFs) in the frequency domain. The Hilbert transform, a mathematical transformation used to construct the analytical form of the signal, can extract indices such as the average energy, energy entropy, and instantaneous frequency variance of each EMF across different frequency bands. The set of all indices is then used as the frequency domain energy distribution characteristic. Secondly, optical flow field calculation is a process of estimating the motion velocity vector of each pixel in an image based on the constraint of pixel brightness changes between adjacent frames in an image sequence. The statistical characteristics of the motion vector field extracted from the image frame sequence of a multi-source heterogeneous monitoring dataset can be obtained using the pyramid Lucas-Cannard algorithm. Then, the Park transform is a mathematical transformation process that converts AC quantities in a two-phase stationary coordinate system to a two-phase rotating coordinate system that rotates synchronously with the rotor magnetic field. This can project the current component onto... By rotating the coordinate system along its direct and quadrature axes, the direct-axis current components and quadrature-axis current components are obtained. In motor control, these components correspond to the currents that generate the magnetic field and the currents that generate torque, respectively. The total harmonic distortion (THD) of the output current signal can also be calculated. THD describes the degree to which the current waveform deviates from a standard sine wave. The set of direct-axis current components, quadrature-axis current components, and THD can be used as the electrical state characteristics. Finally, principal component analysis (PCA) is a statistical process that transforms potentially correlated variables into a set of linearly uncorrelated new variables through orthogonal transformation. The frequency domain energy distribution characteristics, the statistical characteristics of the motion vector field, and the electrical state characteristics can be combined to form a high-dimensional eigenvector. PCA calculates the covariance matrix of this high-dimensional eigenvector, then solves for the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvectors of the principal components are selected in descending order of eigenvalues, and the resulting eigenvector is used as the operating state characteristic vector.

[0029] It should be noted that the operating state feature vector in this application is a low-dimensional digital representation that has undergone in-depth processing and fusion. It is a comprehensive state description index formed by parsing the inherent frequency domain energy structure from the vibration signal through variational mode decomposition and Hilbert transform, quantifying the macroscopic motion law from the image through optical flow field calculation, separating the key electrical control components from the current signal through Park transform, and finally eliminating redundancy and retaining core information through principal component analysis. The operating state feature vector integrates deep state information from three dimensions: mechanical vibration, visual motion, and motor electrical, which can provide efficient and high-quality input. In addition, variational mode decomposition aims to adaptively extract the inherent modes of the vibration signal, which is beneficial to overcome mode aliasing; Park transform aims to decouple the motor current, so that the features can be more directly associated with the torque and magnetic field state; and principal component analysis fusion aims to achieve data dimensionality reduction and decorrelation, which is beneficial to improving computational efficiency and generalization ability.

[0030] In this embodiment, the device health baseline model based on the operating state feature vector can be established in the following manner: A normal operating condition sample set is constructed based on the aforementioned operating state feature vectors; Using a support vector machine with the normal operating condition sample set as training data, find the smallest hypersphere; A health baseline model for the device is constructed based on the center and radius of the minimum hypersphere.

[0031] In specific implementation, firstly, all operating status feature vectors of the automatic cell reversal device during historical periods of good working condition (i.e., no alarms, no abnormal shutdowns) are collected, and all operating status feature vectors are aggregated to form a normal operating condition sample set. Then, using the support vector machine algorithm as a modeling tool, and the normal operating condition sample set as training data, a hypersphere with the smallest volume that can surround all or most normal samples can be found in the high-dimensional feature space composed of operating status feature vectors. This is beneficial for balancing the weight of hypersphere error. Finally, when the training process is completed and converged, the support vector machine algorithm will output the found minimum hypersphere. The minimum hypersphere includes two key parameters: the point located in the high-dimensional feature space is called the center of the sphere, and the length value that determines the size of the sphere is the radius. The mathematical model composed of the center of the sphere and the radius can be used as the equipment health baseline model.

[0032] It should be noted that the equipment health baseline model in this application is a geometric boundary model trained on normal operation data by the support vector machine algorithm. It can define the distribution area of ​​all normal samples in a high-dimensional feature space using the smallest hypersphere. It is used in scenarios where only normal operating condition samples can be obtained. Its goal is to accurately describe the appearance of "health" rather than distinguishing multiple specific fault types. Therefore, it is very suitable for early warning and anomaly detection of equipment status.

[0033] In step S3, the parameters of the device health baseline model are adaptively tuned using a genetic optimization strategy to obtain the state deviation residuals.

[0034] In this embodiment, the parameters of the device health baseline model are adaptively tuned using a genetic optimization strategy to obtain the state deviation residuals. Specifically, this can be achieved in the following manner: The fitness function is determined based on the device health baseline model. When the fitness function converges, the device health baseline model is retrained using the genetic individuals extracted from the support vector machine, thereby obtaining the state update vector; The state deviation residual is determined by the degree of deviation between the state update vector and the center of the smallest hypersphere; The genetic individuals extracted by the support vector machine include the kernel width parameter and the error penalty factor.

[0035] In practical implementation, firstly, the fitness function is used to evaluate the quality of the equipment health baseline model under different parameter settings. The fitness function value can be the weighted sum of the false alarm rate and the model's hypersphere volume on an independent validation sample set. The false alarm rate refers to the proportion of normal samples misclassified as abnormal, and the hypersphere volume is calculated using the radius and describes the model's compactness. Then, a genetic optimization strategy is initiated to adaptively tune the parameters of the equipment health baseline model. This is achieved through the kernel width parameter and error penalty factor of the support vector machine. In genetic optimization, the kernel width parameter and error penalty factor are encoded into a string as a genetic individual. In actual implementation, an initial set of genetic individuals can be randomly generated. The system consists of individuals forming a population. The fitness function value for each genetic individual is calculated. Based on the fitness value, a selection of superior individuals is made from the current population using a roulette wheel selection method. Single-point crossover is performed on the selected individuals to generate new parameter combinations. Then, a bit-flipping mutation operation is performed on the encoding of some individuals with a small probability to introduce randomness. This selection, crossover, and mutation process is iterated repeatedly to obtain a retrained equipment health baseline model. The operating state feature vector in the retrained equipment health baseline model can then be used as the state update vector. Finally, the Euclidean distance from the state update vector to the center of the smallest hypersphere of the retrained equipment health baseline model is calculated, and the obtained distance value is used as the state deviation residual.

[0036] It should be noted that the genetic optimization strategy in this application is a global optimization algorithm that simulates the biological evolution process and is used to automatically search for the optimal model parameters in a class of support vector machines; the kernel function width parameter is an indicator that determines the smoothness of the data distribution in the high-dimensional feature space; the error penalty factor is a measure of the model's tolerance for outliers in the training samples; and the state deviation residual is a scalar value that characterizes the degree of deviation between the current operating state of the equipment and the health baseline model. It can convert the complexity of the equipment state into a measurable degree of deviation and is the core signal driving the entire early warning and maintenance process.

[0037] In step S4, real-time monitoring data of the automatic cell reversal device is acquired, and multi-scale confidence fusion is performed on the state deviation residual based on the real-time monitoring data to generate a comprehensive health index sequence. The comprehensive health index sequence is then classified into anomaly levels based on the entropy weight decision mechanism to obtain the anomaly level classification result.

[0038] In practice, real-time monitoring data of the automatic cell flipping device is acquired through vibration sensors, current sensors, and industrial cameras. It should be noted that the real-time monitoring data in this application refers to the set of raw signals describing the current instantaneous operating state of the automatic cell flipping device, which are synchronously collected from three different types of sensors: vibration, electrical, and vision. These include real-time vibration signals, real-time current signals, and real-time image sequences. The real-time vibration signal is a value representing the dynamic response of the mechanical structure; the real-time current signal describes the instantaneous load of the motor and the electrical state of the drive system; and the real-time image sequence provides visual evidence and spatial location information of the motion process.

[0039] Preferably, in this embodiment, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the comprehensive health index sequence according to the present application. In this embodiment, the comprehensive health index sequence is generated by performing multi-scale confidence fusion on the state deviation residuals based on the real-time monitoring data, which can be achieved through the following steps: First, in step S41, the state deviation residual is subjected to dual-tree complex wavelet transform to obtain sub-band coefficients at different scales; Then, in step S42, the energy entropy of the subband coefficients at each scale is calculated to obtain the local confidence at different scales; Finally, in step S43, the weights of each scale are dynamically allocated according to the local confidence at different scales, and then the comprehensive health index sequence is calculated based on all weights.

[0040] In practical implementation, firstly, the dual-tree complex wavelet transform is a digital filtering method for multi-resolution signal analysis. It consists of two parallel real wavelet transform trees, which process the real and imaginary parts of the signal respectively, thus obtaining coefficient sequences distributed across multiple different resolution levels (i.e., different scales). These coefficient sequences can then be used as sub-band coefficients for the corresponding scale. Secondly, energy entropy is an indicator of the uniformity or complexity of signal energy distribution. The sum of the squares of all sub-band coefficients at a given scale can be calculated as the total energy at that scale. Dividing the square of each sub-band coefficient by the total energy yields the proportion of that coefficient in the total energy, i.e., the probability distribution. Finally, the probability distribution is calculated using the information entropy formula. The entropy value of the rate distribution is obtained, and then the energy entropy at each scale is normalized by Z-score to obtain a value between 0 and 1. This value can be used as the local confidence level to obtain the local confidence level at different scales. Finally, the softmax function can be used to determine the weight of each scale through nonlinear mapping, ensuring that the sum of all weights is one. The local confidence level at each scale is multiplied by its corresponding dynamic weight, and then all weighted results are added together to obtain a comprehensive scalar value. The above process is repeated for the local confidence levels at all scales, and the resulting numerical sequence can be used as the comprehensive health index sequence.

[0041] It should be noted that the multi-scale confidence fusion in this application combines time-frequency analysis with information metric signal processing; the dual-tree complex wavelet transform decomposes a one-dimensional time-series signal into different scales, thereby separating components with different time characteristics, such as transient impacts, periodic fluctuations, and slow drifts, that may be contained in the signal; local confidence is a quantitative assessment of the information quality of a signal component at a certain scale based on energy entropy calculation. The higher the local confidence, the stronger the regularity of the component at that scale, which may correspond to a certain physical process or fault characteristic; the comprehensive health index sequence is a one-dimensional time-series index that integrates multi-scale information. By giving greater weight to high-confidence scales, it enhances the contribution of reliable information and suppresses the influence of noise scales, thus more robustly and sensitively characterizing the overall health status evolution trend of the equipment.

[0042] In this embodiment, the anomaly level classification of the comprehensive health index sequence is based on the entropy weight decision mechanism, and the anomaly level classification result can be obtained in the following specific way: Determine the sample entropy and root mean square of the comprehensive health index sequence; The objective weights of the sample entropy and the root mean square are calculated using the entropy weight method. Construct a membership function based on the sample entropy and the objective weights of the root mean square; The anomaly level classification result is determined by the membership function and the preset anomaly level.

[0043] In practical implementation, firstly, the proportion of similar subsequences in the comprehensive health index sequence is statistically analyzed. Based on this proportion, a non-negative value representing the degree of irregularity of the sequence is calculated using the information entropy formula. This value is used as the sample entropy, and the root mean square (RMS) is calculated using the root mean square formula for the comprehensive health index sequence. Secondly, the entropy weighting method is an objective weighting process based on the concept of information entropy to determine the importance of each indicator in decision-making. The sample entropy and the RMS are normalized to eliminate the influence of dimensions. The value of the normalized sample entropy is then used as the objective weight of the sample entropy, and the value of the normalized RMS is used as the objective weight of the RMS. Then, the membership function is a function in fuzzy logic used to describe the degree to which an element belongs to a certain fuzzy set. It can be a preset anomaly level (e.g., ...). For example, membership functions are constructed for "Normal," "Warning," "Alarm," and "Emergency." This involves setting ideal value ranges for sample entropy and root mean square (RMS) for each anomaly level. Then, using common mathematical functions such as Gaussian or trigonometric functions, the calculated sample entropy and RMS values ​​are mapped to their membership degrees for each anomaly level. The membership degrees of sample entropy and RMS for the same level are multiplied by their respective objective weights calculated in the previous step, and then summed. This calculation process serves as the membership function. Finally, the comprehensive membership degree of the current comprehensive health index sequence to each preset anomaly level is calculated. The comprehensive membership degrees of all anomaly levels are compared, and the level with the highest comprehensive membership degree is selected as the anomaly level classification result.

[0044] It should be noted that the entropy weight decision-making mechanism in this application is an objective and automated decision-making process that combines information entropy theory and fuzzy logic. Sample entropy is used to measure the complexity and unpredictability of the comprehensive health index sequence; a higher sample entropy indicates an abnormal state of the cell automatic reversal device. The root mean square (RMS) is used to measure the energy level of the overall fluctuation of the sequence. Furthermore, the entropy weight method in this application objectively allocates decision weights based on the amount of information actually provided by the sample entropy and RMS in historical data, avoiding bias caused by subjective weight setting. The membership function realizes the mapping relationship from precise values ​​to fuzzy concepts, handling transitions with unclear state boundaries. The anomaly level classification result quantifies the continuously changing health state into an operable early warning level, providing a clear decision-making basis for operation and maintenance scheduling.

[0045] In step S5, the pre-maintenance scheduling and fault location of the automatic cell flipping device are realized by driving the operation and maintenance instructions based on the abnormality level classification results.

[0046] In this embodiment, the pre-maintenance scheduling and fault location of the automatic cell flipping device can be achieved by driving maintenance instructions based on the anomaly level classification results in the following manner: Based on the anomaly level classification results, operation and maintenance instructions are generated, and key scale features and sensor channel information that triggered the anomaly level classification results are traced back synchronously. By combining the key scale features and the sensor channel information, the faulty components and causes are queried to obtain fault location information; The operation and maintenance instructions enable pre-maintenance scheduling of the automatic cell flipping device, and the fault location information enables fault location of the automatic cell flipping device.

[0047] In practice, firstly, based on a pre-defined rule base mapping different anomaly levels to specific maintenance actions, for example, when the anomaly level is classified as "early warning," the corresponding maintenance action mapped by the rule base is "record the status log and prompt the operator to conduct on-site inspection"; when classified as "alarm," the action is "generate and issue a preventive maintenance work order and notify the spare parts warehouse to prepare relevant consumables"; when classified as "emergency," the action is "immediately send an emergency stop command to the device control system and trigger the on-site audible and visual alarm." The system can automatically retrieve and execute the corresponding action from the rule base based on the real-time anomaly level to obtain maintenance instructions. Secondly, it can trace back to the multi-scale confidence fusion step, retrieve the sub-band coefficient feature corresponding to the scale with the highest local confidence when calculating the comprehensive health index, and identify this feature as the key scale feature. Simultaneously, it traces back to real-time monitoring data to analyze the period during which the signal of the specific sensor (vibration, current, image) first appears or experiences the most severe abnormal fluctuation, thereby determining the sensor channel information. Then, the fault knowledge graph is a structured database used to store fault modes, typical characteristics (such as what energy distribution changes occur at what scale), and potentially associated sensor types and physical components. Key scale features and sensor channel information can be used as input conditions to perform matching searches in the fault knowledge graph to obtain fault location information. Finally, the generated maintenance instructions are sent to the controller of the automatic cell flipping device for execution, thus realizing pre-maintenance scheduling. For example, according to the warning level instruction, the controller only records data and illuminates the indicator light; according to the alarm level instruction, it automatically adjusts operating parameters and waits for maintenance; according to the emergency level instruction, it immediately executes a safe shutdown. At the same time, the fault location information is attached to the maintenance work order, which helps to shorten the troubleshooting time.

[0048] It should be noted that the pre-maintenance scheduling in this application refers to proactive and planned maintenance activities based on state prediction and level early warning, before equipment performance deteriorates significantly or serious failures occur, aiming to avoid unplanned downtime; the fault location information is a guiding conclusion about the location and cause of potential faults, derived by combining key features analyzed in real time with historical experience knowledge; in addition, synchronously backtracking key scale features and sensor channel information, and using fault knowledge graphs for matching, helps to improve the pertinence and efficiency of operation and maintenance response.

[0049] In summary, the technical solution adopted in this application can achieve accurate assessment of the operating status of the automatic cell flipping device, early warning of faults, and precise location, thereby improving the operation and maintenance efficiency of battery production equipment.

[0050] Example 2: This application provides a monitoring system for the operating status of an automatic cell flipping device, referring to... Figure 4 As shown, this figure is a modular structure diagram of a cell automatic flipping device operation status monitoring system according to this application. The status monitoring system includes: The data synchronization module 100 is used to collect the operating parameters and image data of the automatic cell flipping device, and then obtain a multi-source heterogeneous monitoring dataset. The feature decoupling module 200 is used to perform time-frequency domain feature decoupling on the multi-source heterogeneous monitoring dataset to obtain the operating state feature vector, and to establish a device health baseline model based on the operating state feature vector. The residual generation module 300 is used to adaptively tune the parameters of the device health baseline model through a genetic optimization strategy to obtain the state deviation residual. The multi-scale entropy weight decision module 400 is used to acquire real-time monitoring data of the automatic cell reversal device, perform multi-scale confidence fusion on the state deviation residual based on the real-time monitoring data, generate a comprehensive health index sequence, and classify the comprehensive health index sequence into anomalies based on the entropy weight decision mechanism to obtain the anomaly classification result. The operation and maintenance scheduling module 500 is used to drive operation and maintenance instructions based on the abnormality level classification results to realize the pre-maintenance scheduling and fault location of the automatic cell flipping device.

[0051] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0052] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0053] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for monitoring the operating status of an automatic cell flipping device, characterized in that, The status monitoring method includes the following steps: The operating parameters and image data of the automatic cell flipping device are collected to obtain a multi-source heterogeneous monitoring dataset; The time-frequency domain features of the multi-source heterogeneous monitoring dataset are decoupled to obtain the operating status feature vector, and a device health baseline model is established based on the operating status feature vector. The parameters of the device health baseline model are adaptively tuned using a genetic optimization strategy to obtain the state deviation residuals. Real-time monitoring data of the automatic cell flipping device is obtained. Based on the real-time monitoring data, multi-scale confidence fusion is performed on the state deviation residual to generate a comprehensive health index sequence. Based on the entropy weight decision mechanism, the comprehensive health index sequence is classified into abnormal levels to obtain the abnormal level classification result. Based on the anomaly level classification results, the operation and maintenance instructions are driven to realize the pre-maintenance scheduling and fault location of the automatic cell flipping device; Specifically, the parameters of the device health baseline model are adaptively tuned using a genetic optimization strategy to obtain the state deviation residuals, which include: The fitness function is determined based on the device health baseline model. When the fitness function converges, the device health baseline model is retrained using the genetic individuals extracted from the support vector machine, thereby obtaining the state update vector; The state deviation residual is determined by the degree of deviation between the state update vector and the center of the smallest hypersphere; The genetic individuals extracted by the support vector machine include a kernel function width parameter and an error penalty factor. The genetic optimization strategy is a global optimization algorithm that simulates the biological evolution process. The kernel function width parameter is an indicator that determines the smoothness of the data distribution in the high-dimensional feature space. The error penalty factor controls the model's tolerance for outliers in the training samples. Specifically, the generation of a comprehensive health index sequence by performing multi-scale confidence fusion on the state deviation residuals based on the real-time monitoring data includes: The state deviation residual is subjected to dual-tree complex wavelet transform to obtain sub-band coefficients at different scales; Calculate the energy entropy of the subband coefficients at each scale to obtain the local confidence at different scales; The weights of each scale are dynamically allocated based on the local confidence level at different scales, and then a comprehensive health index sequence is calculated based on all weights. Specifically, the abnormality level classification of the comprehensive health index sequence based on the entropy weight decision mechanism yields the following results: Determine the sample entropy and root mean square of the comprehensive health index sequence; The objective weights of the sample entropy and the root mean square are calculated using the entropy weight method. Construct a membership function based on the sample entropy and the objective weights of the root mean square; The anomaly level classification result is determined by the membership function and the preset anomaly level.

2. The method for monitoring the operating status of an automatic cell flipping device as described in claim 1, characterized in that, The operation parameters and image data of the automatic cell flipping device are collected to obtain a multi-source heterogeneous monitoring dataset, which specifically includes: The three-phase current signal and vibration acceleration signal of the automatic cell flipping device are collected, and the time-series image frames during the movement of the flipping robotic arm are captured. The three-phase current signal is subjected to Clarke transform to obtain the current component in the two-phase stationary coordinate system. The vibration acceleration signal, the current component and the time sequence image frame are used to form the original dataset. The original dataset is subjected to multi-channel synchronous calibration and timestamp alignment to generate a multi-source heterogeneous monitoring dataset.

3. The method for monitoring the operating status of an automatic cell flipping device as described in claim 1, characterized in that, The time-frequency domain feature decoupling of the multi-source heterogeneous monitoring dataset to obtain the operating state feature vector specifically includes: Variational mode decomposition is performed on the vibration signal components in the multi-source heterogeneous monitoring dataset, and then the frequency domain energy distribution characteristics are obtained by Hilbert transform. Optical flow field calculations are performed on the image frame sequences in the multi-source heterogeneous monitoring dataset to extract statistical features of the motion vector field; The Park transform is performed on the current components in the multi-source heterogeneous monitoring dataset to obtain the electrical state characteristics; The frequency domain energy distribution characteristics, the statistical characteristics of the motion vector field, and the electrical state characteristics are fused by principal component analysis to obtain the operating state feature vector.

4. The method for monitoring the operating status of an automatic cell flipping device as described in claim 1, characterized in that, Establishing a device health baseline model based on the aforementioned operating status feature vector specifically includes: A normal operating condition sample set is constructed based on the aforementioned operating state feature vectors; Using a support vector machine with the normal operating condition sample set as training data, find the smallest hypersphere; A health baseline model for the device is constructed based on the center and radius of the minimum hypersphere.

5. The method for monitoring the operating status of an automatic cell flipping device as described in claim 1, characterized in that, Real-time monitoring data of the automatic cell flipping device is obtained through vibration sensors, current sensors, and industrial cameras.

6. The method for monitoring the operating status of an automatic cell flipping device as described in claim 1, characterized in that, Based on the anomaly level classification results, the pre-maintenance scheduling and fault location of the automatic cell flipping device are driven by operation and maintenance instructions, specifically including: Based on the anomaly level classification results, operation and maintenance instructions are generated, and key scale features and sensor channel information that triggered the anomaly level classification results are traced back synchronously. By combining the key scale features and the sensor channel information, the faulty components and causes are queried to obtain fault location information; The operation and maintenance instructions enable pre-maintenance scheduling of the automatic cell flipping device, and the fault location information enables fault location of the automatic cell flipping device.

7. A battery cell automatic flipping device operation status monitoring system, used to execute the battery cell automatic flipping device operation status monitoring method as described in any one of claims 1 to 6, characterized in that, The status monitoring system includes: The data synchronization module is used to collect the operating parameters and image data of the automatic cell flipping device, thereby obtaining a multi-source heterogeneous monitoring dataset. The feature decoupling module is used to decouple the time-frequency domain features of the multi-source heterogeneous monitoring dataset to obtain the operating state feature vector, and to establish a device health baseline model based on the operating state feature vector. The residual generation module is used to adaptively tune the parameters of the device health baseline model through a genetic optimization strategy to obtain the state deviation residual. The multi-scale entropy weight decision module is used to acquire real-time monitoring data of the automatic cell reversal device, perform multi-scale confidence fusion on the state deviation residual based on the real-time monitoring data, generate a comprehensive health index sequence, and classify the comprehensive health index sequence into anomaly levels based on the entropy weight decision mechanism to obtain the anomaly level classification result. The operation and maintenance scheduling module is used to drive operation and maintenance instructions based on the anomaly level classification results to realize the pre-maintenance scheduling and fault location of the automatic cell flipping device.

Images