Battery part production traceability management system based on Internet of Things

Abstract

The invention belongs to the technical field of intelligent manufacturing traceability, and provides a battery part production traceability management system based on the Internet of Things, which comprises the steps of collecting battery part information, setting double identifiers for each battery part, storing the battery part information, calculating a qualified index of the battery part for preliminary evaluation, and determining whether the battery part is qualified or not. And for the preliminarily qualified battery parts, constructing and training an attenuation prediction model to predict the service lives of the battery parts to obtain the predicted actual service lives of the battery parts, dividing the battery parts into inferior products, non-defective products and superior products, constructing and training a defect analysis model to identify the defect root causes of the inferior products, and determining the actual service lives of the battery parts. The method comprises the following steps: determining steps causing defect root causes, setting risk mark values, recording the number of times of causing the defect root causes in a batch, sorting each step in a production line, obtaining a tracing priority sequence, calculating the inferior product rate of battery parts in the batch, and judging whether to trigger forward recall or reverse tracing.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing traceability technology, specifically an Internet of Things-based traceability management system for battery component production. Background Technology

[0002] Against the backdrop of the rapid development of the new energy battery industry, the quality of battery components directly determines the safety, reliability and service life of the battery. However, the existing traceability management model for battery component production has many technical bottlenecks and is difficult to meet the requirements of high-precision, full-process and traceable quality control.

[0003] In the existing production system, information collection for battery components is mostly done manually or through single-device sensing, resulting in data fragmentation and poor timeliness. For example, production process data is only recorded through local equipment, making it impossible to achieve time-series correlation of the entire process, such as material preparation, molding, and sintering. Product quality inspection data lacks real-time binding with production process data, leading to information gaps when tracing quality issues. In terms of identification technology, traditional barcodes or QR codes can only store simple batch information and are easily worn and copied, making it difficult to achieve unique identification of components throughout their entire lifecycle. Some companies have tried to use RFID technology, but they have not formed a dual verification mechanism combining explicit and implicit codes, posing a risk of information forgery and failing to ensure the integrity and immutability of process parameters.

[0004] On the other hand, existing quality assessments lack evaluation of the long-term reliability of components. For example, aging degradation characteristics are not included in the quality grading standards, resulting in some components that meet the initial indicators but have poor anti-aging capabilities entering the market, increasing the risk of battery failure in the later stages. When defective products appear, existing systems rely heavily on manual experience to judge the root cause of defects, lacking data-driven intelligent analysis methods. Since the mapping relationship between the root cause of defects and production steps has not been established, it is difficult to quickly locate the problematic link, leading to the repeated occurrence of similar defects. In batch risk management, risk is judged only by the single indicator of defect rate, without combining the frequency of problems in each production step for comprehensive evaluation, which easily leads to misjudgment or omission of risks. In the traceability process, there is a lack of priority ranking between forward recall and reverse tracing, resulting in low recall efficiency and excessively long problem investigation time.

[0005] To address the above problems, this invention proposes an Internet of Things-based traceability management system for battery component production. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve the technical problem is: a battery component production traceability management system based on the Internet of Things, comprising:

[0008] Data Acquisition and Identification Module: Collects information on battery components, sets dual identification for each battery component including an internal RFID tag and a surface laser QR code, and stores the information on the battery components;

[0009] Quality assessment module: Based on the collected information on battery components, it determines whether the battery components are initially qualified, builds and trains a degradation prediction model to predict the lifespan of the initially qualified battery components, obtains the predicted actual lifespan of the battery components, and classifies the battery components into inferior, good and superior products based on the initial qualification judgment results.

[0010] Defect Analysis Module: Constructs and trains defect analysis models to identify the root causes of defects in defective products, determines the steps in the production line that cause the root causes of defects, and sets risk flag values ​​for each step to quantify the risk of the step.

[0011] Defect traceability module: Based on the risk flag value, each step in the production line is sorted to obtain a traceability priority sequence. Combined with the calculated defect rate of battery parts in the batch, the batch risk is assessed. If the risk is high, a positive recall is triggered. If the unrecalled battery parts have defect problems, a reverse traceability is triggered.

[0012] The method for storing battery component information is as follows:

[0013] Battery component information includes process parameters and basic data of battery components. The battery component information is divided into sub-datasets, including basic battery component dataset, raw material dataset, production process dataset, and product quality inspection dataset. The product quality inspection dataset includes basic test data group and deep test data group. After data verification, the data is stored in a distributed database to obtain the battery component database.

[0014] Each battery component is assigned an identifier that includes both a coded and a plaintext code. The coded code is set as an internal RFID tag that stores the hash value of the process parameters, while the plaintext code is set as a surface laser QR code that stores the basic data of the battery component. The surface laser QR code is matched one-to-one with the internal RFID tag, and the association is achieved through a database association table for the battery components.

[0015] The method for determining whether the battery components are initially qualified is as follows:

[0016] Obtain the product quality inspection dataset of battery components, set the basic evaluation conditions for battery components, including the allowable range of test values ​​for each data item in the basic test data set, compare the basic test data set of battery components with the basic evaluation conditions, calculate the pass index of battery components based on the indicator function, if the pass index Q is less than 1, the battery component is judged to be unqualified, if the pass index Q is equal to 1, the battery component is judged to be initially qualified.

[0017] The method for classifying battery components is as follows:

[0018] If the battery components are initially qualified, obtain the predicted actual lifespan of the battery components. If it is greater than or equal to the preset excellent product lifespan standard, the battery components are judged as excellent products. If it is less than the excellent product lifespan standard, but greater than or equal to the good product lifespan standard, the battery components are judged as good products.

[0019] If battery components are substandard, or if the predicted actual lifespan of battery components is less than the standard lifespan of good products, the battery components are judged to be defective.

[0020] The method for obtaining the predicted actual lifespan is as follows:

[0021] Obtain a deep test data set for battery components. The deep test data set includes an aging acceleration factor and an aging decay sequence. The aging decay sequence includes the performance retention rate of the battery components at each performance test moment during the aging test. Normalize the aging decay sequence to obtain a normalized dataset.

[0022] A degradation prediction model is constructed and trained. The trained degradation prediction model is used to iteratively predict the performance retention rate of battery components until the predicted performance retention rate is less than the preset retention threshold. At this point, it is determined that the battery components can no longer be used. The duration between the end point of the normalized dataset and the time when the battery components can no longer be used is calculated to obtain the test life of the battery components. The data is then processed in conjunction with the aging acceleration factor to calculate the predicted actual life of the battery components.

[0023] The decay prediction model is constructed and trained as follows:

[0024] An LSTM network was used to construct a decay prediction model, which included an input layer, two LSTM layers, a dropout layer, and an output layer. Each LSTM layer had 64 neurons and ReLU was used as the activation function.

[0025] Obtain and divide the normalized dataset into training and validation sets in an 8:2 ratio. Input the training set into the decay prediction model to train the decay prediction model. Use the mean squared error as the loss function, select the Adam optimizer, and iteratively adjust the parameters until the loss function of the decay prediction model no longer changes in the validation set, thus obtaining the trained decay prediction model.

[0026] The risk indicator value is obtained in the following way:

[0027] Establish a mapping table between the root causes of defects and each step of the production line. For the root causes of defects of the identified defective products, determine the steps in the production line that caused the root causes of defects. When each batch of battery components starts production, set a risk flag value for each step and assign it an initial value of 0. For the steps mapped to the root causes of defects of the identified defective products, increment the risk flag value corresponding to the step.

[0028] The method for identifying the root causes of defects is as follows:

[0029] If the battery components are defective, mark them as defective products to be analyzed. Set a historical time period, obtain all defective products with the same production line number as the defective products to be analyzed within the historical time period and integrate them into a historical defective product set. Mark the defective products with the corresponding root causes of defects, obtain and associate them with the process feature matrix of the defective products, and integrate them to obtain a training sample set.

[0030] A defect analysis model is constructed using a convolutional neural network, which includes an input layer, three convolutional layers, a global average pooling layer, and an output layer. The input layer receives the process feature matrix, and the output layer outputs the probability distribution of the root causes of defects. The defect analysis model is trained using a training sample set to obtain the trained defect analysis model.

[0031] The process feature matrix of the defective product to be analyzed is input into the trained defect analysis model to obtain the probability distribution of each defect root cause of the defective product to be analyzed. If the maximum probability value is less than the preset probability threshold, manual review is triggered to determine the defect root cause of the defective product to be analyzed; otherwise, the defect root cause of the defective product to be analyzed is determined to be the defect root cause corresponding to the maximum probability value.

[0032] The process feature matrix is ​​obtained as follows:

[0033] Obtain battery component information for each defective product in the defective product set to be analyzed and the historical defective product set. The battery component information includes process parameters. For data with time-series characteristics in the process parameters, extract time-series characteristics including mean, peak value and volatility, and replace the corresponding data in the process parameters with the extracted time-series characteristics. In the historical defective product set, z-score standardization is used to standardize each data of the replaced process parameters, and the process feature matrix of each defective product is integrated.

[0034] The method for assessing batch risk is as follows:

[0035] For any batch of battery components, the risk indicator value of each step in the corresponding production line is obtained in real time, and each step is sorted from largest to smallest based on the risk indicator value to obtain a traceability priority sequence. The number of defective battery components in the batch is obtained in real time for data processing, and the defect rate of battery components in the batch is calculated. If the maximum value of the risk indicator value in the traceability priority sequence is greater than the preset risk threshold, or the defect rate of battery components in the batch is greater than the preset defect rate threshold, the batch is judged to be of high risk.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. This invention sets dual identifiers for each battery component through a data acquisition and identification module, enabling accurate data collection and storage, laying the foundation for subsequent management. The quality assessment module can scientifically classify battery components into different grades, identify potential defective products in advance, and prevent them from entering the market and causing losses. The defect analysis module accurately locates the root cause of defects and corresponding production steps, and quantifies the risks of each step, helping companies focus on key issues, improve processes in a targeted manner, and enhance product quality and production efficiency.

[0038] 2. This invention generates a traceability priority sequence based on risk indicator values ​​and determines whether a positive recall is triggered by combining the defect rate. This allows for rapid identification of the source of production problems and timely measures to reduce losses. For non-recalled battery components, it can determine whether reverse traceability is triggered. Once a problem occurs, it can be quickly traced back to the production process. At the same time, the complete management process helps enterprises optimize production management and enhance market competitiveness. Attached Figure Description

[0039] The invention will now be further described with reference to the accompanying drawings.

[0040] Figure 1 This is a system module architecture diagram of a battery component production traceability management system based on the Internet of Things, as described in an embodiment of the present invention.

[0041] Figure 2 This is a flowchart illustrating the steps of a battery component production traceability management system based on the Internet of Things, as described in an embodiment of the present invention. Detailed Implementation

[0042] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0043] Example 1

[0044] Please see Figure 1 As shown in the embodiment of the present invention, a battery component production traceability management system based on the Internet of Things includes the following modules:

[0045] Data Acquisition and Identification Module: Collects information on battery components, sets dual identification for each battery component including an internal RFID tag and a surface laser QR code, and stores the information on the battery components;

[0046] An IoT distributed acquisition architecture is used to collect information on battery components. The information includes process parameters and basic data of battery components. The process parameters include raw material data, production process data and product quality inspection data. The basic data of battery components includes batch number, production time, battery component serial number and production line number. The basic data of battery components is integrated to obtain the basic dataset D1 of battery components.

[0047] The raw material data includes raw material composition, purity, and storage condition parameters. The raw material composition includes the content of key elements such as lithium, cobalt, and nickel. The purity represents the content of the main components and is obtained through the raw material quality inspection reports uploaded by suppliers to the blockchain platform. The storage condition parameters include storage temperature and storage humidity, which are collected through warehouse environmental sensors. The raw material data is integrated to obtain the raw material dataset D2.

[0048] The production process data covers four key processes: batching, molding, sintering, and packaging. For each key process, the production process data includes processing time and ambient humidity. Processing time represents the actual time consumed by each key process and is collected through a process timer. Ambient humidity represents the ambient humidity of the processing workshop and is collected through an environmental sensor. For the key process of molding, the production process data also includes processing pressure, which represents the stamping pressure value during molding and is collected through a pressure sensor. For the key process of sintering, the production process data also includes the process temperature curve, which represents the real-time temperature change during sintering and is collected through a furnace temperature tracker. The production process data is integrated according to the time sequence of each key process to obtain the production process dataset D3.

[0049] Among them, product quality inspection data includes dimensional accuracy. ,hardness Salt spray corrosion rate and appearance defects Dimensional accuracy represents the deviation between the actual dimensions and design dimensions of battery components. Actual dimensions are measured using a coordinate measuring machine. Hardness is obtained by averaging multiple measurements using a Brinell hardness tester. Salt spray corrosion rate is obtained by detecting and processing the deviation in corrosion degree of Si'anchi battery components before and after salt spray testing. Appearance defects are identified using a machine vision inspection system to detect scratches, deformations, and other defects, and the identified defects are quantified using the defect area ratio. The dimensional accuracy, hardness, corrosion resistance, and appearance defects in the product quality inspection data are integrated to obtain the basic test data set. i = 1, 2, 3, 4;

[0050] The product quality inspection data also includes aging acceleration factors and aging degradation sequences. The aging acceleration factor represents the acceleration multiple of the test duration in the aging test to the equivalent actual usage time. It is obtained by collecting and processing aging test parameters during the aging test, including aging test parameters such as aging temperature, aging humidity, and aging duration. The aging degradation sequence includes the performance retention rate of battery components at each performance test moment in the aging test. The performance retention rate at each performance test moment is integrated according to the time sequence to obtain the aging degradation sequence. The interval between adjacent performance test moments in the aging test is the same. The aging acceleration factor and aging degradation sequence in the product quality inspection data are integrated to obtain the in-depth test data set.

[0051] The basic test data set and the in-depth test data set of product quality inspection data are integrated to obtain the product quality inspection dataset D4;

[0052] Each battery component is identified by both a coded and a plaintext code. The coded code is an internal RFID tag, which uses high-frequency RFID technology and is embedded inside the battery component during the manufacturing process. The internal RFID tag stores the hash value of the process parameters. The hash value is used to ensure the integrity and immutability of the process parameters and to uniquely identify the production process of the battery component. The plaintext code is a surface laser QR code, which is generated on the surface of the battery component using laser engraving technology. The surface laser QR code stores the basic data of the battery component, which facilitates the quick reading of the basic batch information of the battery component and enables the initial identification and traceability of the battery component.

[0053] For the battery component information, including the basic battery component dataset D1, raw material dataset D2, production process dataset D3, and product quality inspection dataset D4, D1 to D4 are treated as sub-datasets. After data verification, they are stored in a distributed database to obtain the battery component database. Each battery component is identified and bound, and the surface laser QR code is matched one-to-one with the internal RFID tag. The association is achieved through the database association table.

[0054] It should be noted that the function of this module is to collect full lifecycle data of battery components through an IoT distributed architecture, build a battery component database containing four subsets, and achieve dual identification binding through internal RFID tag cryptographic codes and surface laser QR code explicit codes, providing a data foundation and identification support for subsequent traceability management. Data collection covers the entire process of raw materials, production process, and quality inspection. Combined with blockchain to obtain supplier data, it ensures the credibility and integrity of the data source. The dual identification system achieves the dual functions of anti-counterfeiting and rapid traceability. The distributed database and relational table design supports efficient data query and correlation analysis. By integrating IoT distributed collection, blockchain evidence storage and dual identification technologies, it builds an integrated traceability infrastructure for data collection, identification and storage, solving the problems of data fragmentation and easy tampering in traditional traceability.

[0055] Quality assessment module: Based on the collected information on battery components, calculate the qualification index of the battery components for preliminary assessment. For the battery components that are preliminarily qualified, build and train a degradation prediction model to predict the life of the battery components and obtain the predicted actual life of the battery components. Combining the preliminary assessment results and the predicted actual life, the battery components are classified into inferior, good and superior products.

[0056] For battery components, obtain the product quality inspection dataset of battery components, and make a preliminary qualification judgment on the battery components based on the basic test data group X in the product quality inspection dataset.

[0057] Specifically, based on the design standards and industry specifications for battery components, basic evaluation conditions for battery components are set, including the allowable range of test values ​​for each data item in basic test data set X. , The allowable range of test values ​​is ;

[0058] By comparing the basic test data set with the basic evaluation conditions, the pass index Q of the battery components is calculated using the following formula:

[0059] ;

[0060] Where n represents the number of data items in the basic test data set, n=4, and I is an indicator function, when When I=1, it means data Qualified; otherwise, I=0.

[0061] If the calculated pass index Q of the battery component is equal to 1, the basic test data set is considered to be qualified, and the battery component is considered to be initially qualified. If the pass index Q is less than 1, the basic test data set contains unqualified data, the battery component is assessed as unqualified, and the battery component is considered to be a defective product.

[0062] For battery components that have passed preliminary qualification, a degradation prediction model is constructed and trained based on the deep test data set in the product quality inspection dataset of battery components to predict the lifespan of the battery components.

[0063] Specifically, an LSTM network is used to construct a decay prediction model, which includes an input layer, two LSTM layers, a dropout layer, and an output layer. Each LSTM layer has 64 neurons and ReLU is used as the activation function. The dropout layer is used to prevent overfitting.

[0064] Obtain the aging degradation sequence from the deep test data set. The aging degradation sequence includes the performance retention rate of battery components at each performance test moment during the aging test. Normalize the aging degradation sequence to obtain a normalized dataset. Divide the normalized dataset into a training set and a validation set in an 8:2 ratio. Input the training set into the degradation prediction model to train the degradation prediction model. Use the mean squared error as the loss function, select the Adam optimizer, and iteratively adjust the parameters until the loss of the degradation prediction model no longer changes in the validation set. The trained degradation prediction model is then obtained.

[0065] Mark the time corresponding to the end point of the normalized dataset as the cutoff time. Extract a normalized input window sequence from the normalized dataset with the cutoff time as the end point. Input the normalized input window sequence into the trained decay prediction model. Output a normalized prediction window sequence with the same length as the normalized input window sequence, starting from the cutoff time. Perform inverse normalization on the normalized prediction window sequence to obtain the prediction window sequence and the prediction performance retention rate included in the prediction window sequence.

[0066] Obtain the prediction performance retention rate at the end of the prediction window sequence. If the prediction performance retention rate at the end of the prediction window sequence is greater than or equal to the preset retention threshold, it is determined that the battery components can still be used. Input the normalized prediction window sequence into the trained attenuation prediction model to continue prediction and obtain a new prediction window sequence.

[0067] If the prediction performance retention rate at the end of the prediction window sequence is less than the preset retention threshold, it is determined that the battery component can no longer be used, the prediction is stopped, and based on all the prediction window sequences obtained, the duration between the time corresponding to the end of the normalized dataset and the time when the battery component can no longer be used is calculated to obtain the test life of the battery component.

[0068] Obtain the aging acceleration factor from the deep test data set, multiply the test life of the battery components by the aging acceleration factor, and obtain the predicted actual life of the battery components.

[0069] The predicted actual lifespan of battery components is obtained through prediction and calculation, and compared with the preset good product lifespan standard and excellent product lifespan standard respectively. The good product lifespan standard is shorter than the excellent product lifespan standard.

[0070] If the predicted actual lifespan of a battery component is greater than or equal to the standard lifespan of a superior product, the battery component is judged to be a superior product.

[0071] If the predicted actual lifespan of a battery component is less than the standard lifespan of a superior product, but greater than or equal to the standard lifespan of a good product, the battery component is judged to be a good product.

[0072] If the predicted actual lifespan of a battery component is less than the standard lifespan of a good product, the battery component is judged to be a defective product.

[0073] It should be noted that the function of this module is to complete the initial qualification screening based on product quality inspection data, and to predict and classify the life of the initially qualified products by combining the attenuation prediction model built with LSTM network, so as to realize the dynamic evaluation and hierarchical management of product quality. The qualification index standardizes the initial qualification judgment through quantitative indicators, avoiding the subjectivity of human judgment. The attenuation prediction model built with LSTM network predicts the life based on aging attenuation sequence, and converts the actual life into life by combining aging acceleration factor, thereby improving the timeliness and accuracy of life prediction. A two-level evaluation system combining qualification index and time series prediction is proposed. First, basic qualified products are screened through static indicators, and then life is predicted through dynamic time series data, so as to realize the dual evaluation of basic qualification and long-term reliability.

[0074] Defect Analysis Module: Constructs and trains a defect analysis model to identify the root causes of defects in defective products, establishes a mapping table between the root causes of defects and each step of the production line, determines the steps that cause the root causes of defects, and sets risk flag values ​​to record the number of times each step causes the root causes of defects within a batch.

[0075] If the battery parts are determined to be defective, they are marked as defective products to be analyzed. The production line number in the basic data of the battery parts to be analyzed is obtained. A historical time period is set, and all defective products with the same production line number as the defective product to be analyzed within the historical time period are obtained and integrated to obtain the historical defective product set.

[0076] For the defect root cause R corresponding to the defective product label in the historical defective product set, obtain the process parameters of each defective product in the defective product to be analyzed and the historical defective product set, including the raw material dataset, production process dataset, and product quality inspection dataset of the defective product. For the data with time-series features in the process parameters, extract the time-series features including mean, peak value, and volatility, and replace the data with time-series features in the process parameters with the corresponding time-series features. In the historical defective product set, z-score standardization is used to standardize each data of the replaced process parameters, and the process feature matrix of each defective product is integrated.

[0077] For all defective products in the historical defective product set, the process feature matrix of the defective products is associated with the root cause of the defects using the battery component number of the defective product as the key, and the training sample set is obtained.

[0078] A defect analysis model is constructed using a convolutional neural network, comprising an input layer, three convolutional layers, a global average pooling layer, and an output layer. The input layer receives the process feature matrix. The first convolutional layer performs local feature extraction on the input process feature matrix. The second convolutional layer deepens the feature extraction. The third convolutional layer compresses the channel dimension and fuses features, outputting a feature map. The global average pooling layer compresses the feature map into a feature vector. The output layer uses the Softmax activation function to output the probability distribution of the root cause of the defect.

[0079] The defect analysis model is trained using a training sample set, with cross-entropy loss as the loss function, Adam as the optimizer, and a Dropout layer added after the convolutional layer. Combined with an early stopping mechanism, the parameters are iteratively adjusted until the cross-entropy loss of the defect analysis model no longer changes, or until the preset number of iterations is reached, at which point the defect analysis model is judged to have completed training.

[0080] For a defective product to be analyzed, the process feature matrix of the defective product is input into the trained defect analysis model to identify the root cause of the defect. The model outputs the probability distribution of each root cause of the defect and compares the maximum probability value with a preset probability threshold. If the maximum probability value is less than the probability threshold, it is determined that the identification has failed and manual review is triggered to determine the root cause of the defect. Otherwise, the root cause of the defect is determined to be the root cause corresponding to the maximum probability value.

[0081] Establish a mapping table between defect root causes and each step of the production line. For the identified defect root causes, determine the steps in the production line that caused the defect root causes. When each batch of battery components starts production, set a risk flag value for each step and assign it an initial value of 0. For the steps mapped to the defect root causes of the identified defective products, increment the risk flag value corresponding to the step.

[0082] It should be noted that this module is designed for defective products. Based on historical data, it trains a defect analysis model using a convolutional neural network to automatically identify the root causes of defects and mark the corresponding production steps. This enables precise location of the causes of defective products and quantitative tracking of defective steps. The convolutional neural network extracts deep features from the process feature matrix, solving the problem that traditional manual analysis struggles to capture implicit correlations between parameters. The mapping and marking of defect root causes to production steps provides quantitative indicators for subsequent risk assessment, facilitating targeted improvements. The manual review mechanism balances automation efficiency with judgment accuracy, transforming process parameters into a two-dimensional feature matrix. Utilizing the local feature extraction capabilities of the convolutional neural network, it achieves intelligent and automated identification of defect root causes. Risk marker values ​​are introduced to quantify and accumulate step risks, providing traceable quantitative evidence for batch risk assessment and realizing a closed loop of tracing, marking, and improving quality issues.

[0083] Defect tracing module: Based on the risk flag value, each step in the production line is sorted to obtain the traceability priority sequence, the defect rate of battery parts in the batch is calculated, and the traceability priority sequence is used to determine whether to trigger a positive recall. For battery parts that are not recalled, it is determined whether to trigger reverse tracing.

[0084] For any batch of battery components, the risk indicator value of each step in the corresponding production line is obtained in real time, and each step is sorted from largest to smallest based on the risk indicator value to obtain a traceability priority sequence;

[0085] Real-time data on the number of defective battery components within a batch up to the current moment. Calculate the defect rate of battery components within a batch. ;

[0086] ;

[0087] in, This indicates the total planned production quantity of battery components within the batch;

[0088] Risk assessment of batches is conducted based on traceability priority sequences and the defect rate of battery components within the batch.

[0089] Specifically, if the maximum value of the risk flag in the priority sequence is greater than the preset risk threshold, or the defect rate of battery components in a batch is greater than the preset defect rate threshold, the batch is judged to be high-risk, the batch is marked as a risk batch, and a positive recall is triggered.

[0090] If a positive recall is triggered, all non-premium products in the risk batch will be traced and recalled, and a recall list will be generated. Non-premium products include good and inferior products. After the recall is completed, the information will be verified by the surface laser QR code and the internal RFID tag to ensure that the recall targets are accurate. The purpose is to prevent non-premium battery parts in the risk batch from entering the market, and to check and adjust each step of the production line in sequence according to the traceability priority.

[0091] For unrecalled battery parts, if a defect is found in the battery parts, reverse tracing is triggered. The battery part information is read by the surface laser QR code and the internal RFID tag. The batch is traced according to the batch number in the battery part information. The root cause of the defect is investigated in each step of the production line according to the traceability priority sequence of the batch to determine the root cause of the defect in the battery parts.

[0092] It should be noted that this module's function is to identify risky batches based on the defect rate and step-by-step risk indicator value, triggering positive recalls and step adjustments. For defective products, it triggers reverse tracing, investigates root causes according to step-by-step risk priority, and achieves full-chain risk control. The dual-threshold risk assessment of defect rate and risk indicator value takes into account both overall batch quality and local step-by-step risks, reducing risk omissions. Positive recalls prevent non-superior products from entering the market, while reverse tracing investigates according to risk priority, improving the efficiency of defect location. The dual-identification verification during recalls ensures accurate targeting and avoids false recalls. It proposes a two-way traceability mechanism of positive recall and reverse tracing, focusing on risk diffusion prevention and control in the positive direction and on defect root cause location in the reverse direction, forming a closed loop of risk response across the entire chain. Based on the investigation logic of traceability priority sequence, it combines historical defect frequency with real-time risk to improve the pertinence and efficiency of defect analysis.

[0093] The technical solution of this invention is as follows: Information on battery components is collected; each battery component is equipped with a dual identifier including an internal RFID tag and a surface laser QR code; the information on the battery components is stored; based on the collected information, a qualification index is calculated for preliminary evaluation; for preliminarily qualified battery components, a degradation prediction model is constructed and trained to predict the lifespan of the battery components, obtaining the predicted actual lifespan; combining the preliminary evaluation results and the predicted actual lifespan, the battery components are classified into inferior, good, and superior products; a defect analysis model is constructed and trained to identify the root causes of defects in inferior products; a mapping table between the root causes of defects and each step of the production line is established; the steps causing the root causes of defects are determined; and a risk flag value is set to record the number of times each step causes the root causes of defects within a batch; based on the risk flag value, each step in the production line is sorted to obtain a traceability priority sequence; the defect rate of battery components within a batch is calculated; and based on the traceability priority sequence, it is determined whether a positive recall is triggered; for battery components that are not recalled, it is determined whether a reverse traceability is triggered.

[0094] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A battery component production traceability management system based on the Internet of Things, characterized in that: include: Data Acquisition and Identification Module: Collects information on battery components, sets dual identification for each battery component including an internal RFID tag and a surface laser QR code, and stores the information on the battery components; Quality assessment module: Based on the collected information on battery components, it determines whether the battery components are initially qualified, builds and trains a degradation prediction model to predict the lifespan of the initially qualified battery components, obtains the predicted actual lifespan of the battery components, and classifies the battery components into inferior, good and superior products based on the initial qualification judgment results. Defect Analysis Module: Constructs and trains defect analysis models to identify the root causes of defects in defective products, determines the steps in the production line that cause the root causes of defects, and sets risk flag values ​​for each step to quantify the risk of the step. Defect tracing module: Based on the risk indicator value, each step in the production line is sorted to obtain a traceability priority sequence. Combined with the calculated defect rate of battery components within the batch, the batch risk is assessed. If the risk is high, a positive recall is triggered. If the unrecalled battery components have defect problems, a reverse tracing is triggered.

2. The battery component production traceability management system based on the Internet of Things as described in claim 1, characterized in that: The method for storing battery component information is as follows: Battery component information includes process parameters and basic data of battery components. The battery component information is divided into sub-datasets, including basic battery component dataset, raw material dataset, production process dataset, and product quality inspection dataset. The product quality inspection dataset includes basic test data group and deep test data group. After data verification, the data is stored in a distributed database to obtain the battery component database. Each battery component is assigned an identifier that includes both a coded and a plaintext code. The coded code is set as an internal RFID tag that stores the hash value of the process parameters, while the plaintext code is set as a surface laser QR code that stores the basic data of the battery component. The surface laser QR code is matched one-to-one with the internal RFID tag, and the association is achieved through a database association table for the battery components.

3. The battery component production traceability management system based on the Internet of Things according to claim 2, characterized in that: The method for determining whether the battery components are initially qualified is as follows: Obtain the product quality inspection dataset of battery components, set the basic evaluation conditions for battery components, including the allowable range of test values ​​for each data item in the basic test data set, compare the basic test data set of battery components with the basic evaluation conditions, calculate the pass index of battery components based on the indicator function, if the pass index Q is less than 1, the battery component is judged to be unqualified, if the pass index Q is equal to 1, the battery component is judged to be initially qualified.

4. The battery component production traceability management system based on the Internet of Things according to claim 3, characterized in that: The method for classifying battery components is as follows: If the battery components are initially qualified, obtain the predicted actual lifespan of the battery components. If it is greater than or equal to the preset excellent product lifespan standard, the battery components are judged as excellent products. If it is less than the excellent product lifespan standard, but greater than or equal to the good product lifespan standard, the battery components are judged as good products. If battery components are substandard, or if the predicted actual lifespan of battery components is less than the standard lifespan of good products, the battery components are judged to be defective.

5. The battery component production traceability management system based on the Internet of Things according to claim 2, characterized in that: The method for obtaining the predicted actual lifespan is as follows: Obtain a deep test data set for battery components. The deep test data set includes an aging acceleration factor and an aging decay sequence. The aging decay sequence includes the performance retention rate of the battery components at each performance test moment during the aging test. Normalize the aging decay sequence to obtain a normalized dataset. A degradation prediction model is constructed and trained. The trained degradation prediction model is used to iteratively predict the performance retention rate of battery components until the predicted performance retention rate is less than a preset retention threshold. At this point, the battery components are determined to be unusable. The duration between the endpoint of the normalized dataset and the moment when the battery components can no longer be used is calculated to obtain the test lifespan of the battery components. The data is then processed in conjunction with the aging acceleration factor to calculate the predicted actual lifespan of the battery components.

6. The battery component production traceability management system based on the Internet of Things according to claim 5, characterized in that: The decay prediction model is constructed and trained as follows: An LSTM network was used to construct a decay prediction model, which included an input layer, two LSTM layers, a dropout layer, and an output layer. Each LSTM layer had 64 neurons and ReLU was used as the activation function. Obtain and divide the normalized dataset into training and validation sets in an 8:2 ratio. Input the training set into the decay prediction model to train the decay prediction model. Use the mean squared error as the loss function, select the Adam optimizer, and iteratively adjust the parameters until the loss function of the decay prediction model no longer changes in the validation set, thus obtaining the trained decay prediction model.

7. The battery component production traceability management system based on the Internet of Things according to claim 1, characterized in that: The risk indicator value is obtained in the following way: Establish a mapping table between defect root causes and each step of the production line. For the defect root causes of the identified defective products, determine the steps in the production line that caused the defect root causes. When each batch of battery components starts production, set a risk flag value for each step and assign it an initial value of 0. For the steps mapped to the defect root causes of the identified defective products, increment the risk flag value corresponding to the step.

8. The battery component production traceability management system based on the Internet of Things according to claim 7, characterized in that: The method for identifying the root cause of the defect is as follows: If the battery components are defective, mark them as defective products to be analyzed. Set a historical time period, obtain all defective products with the same production line number as the defective products to be analyzed within the historical time period and integrate them into a historical defective product set. Mark the defective products with the corresponding root causes of defects, obtain and associate them with the process feature matrix of the defective products, and integrate them to obtain a training sample set. A defect analysis model is constructed using a convolutional neural network, which includes an input layer, three convolutional layers, a global average pooling layer, and an output layer. The input layer receives the process feature matrix, and the output layer outputs the probability distribution of the root causes of defects. The defect analysis model is trained using a training sample set to obtain the trained defect analysis model. The process feature matrix of the defective product to be analyzed is input into the trained defect analysis model to obtain the probability distribution of each defect root cause of the defective product to be analyzed. If the maximum probability value is less than the preset probability threshold, manual review is triggered to determine the defect root cause of the defective product to be analyzed; otherwise, the defect root cause of the defective product to be analyzed is determined to be the defect root cause corresponding to the maximum probability value.

9. A battery component production traceability management system based on the Internet of Things as described in claim 8, characterized in that: The process feature matrix is ​​obtained as follows: The process involves obtaining battery component information for each defective product in the target defective product set and the historical defective product set. This information includes process parameters. For process parameters with time-series characteristics, time-series features including mean, peak value, and volatility are extracted. The extracted time-series features are then used to replace the corresponding data in the process parameters. Within the historical defective product set, z-score standardization is used to standardize each data point of the replaced process parameters. The results are then integrated to obtain the process feature matrix for each defective product.

10. A battery component production traceability management system based on the Internet of Things according to claim 1, characterized in that: The method for assessing batch risk is as follows: For any batch of battery components, the risk indicator value of each step in the corresponding production line is obtained in real time, and each step is sorted from largest to smallest based on the risk indicator value to obtain a traceability priority sequence. The number of defective battery components in the batch is obtained in real time for data processing, and the defect rate of battery components in the batch is calculated. If the maximum value of the risk indicator value in the traceability priority sequence is greater than the preset risk threshold, or the defect rate of battery components in the batch is greater than the preset defect rate threshold, the batch is judged to be of high risk.

Images