Automobile assembly self-learning method, system, storage medium and electronic device

Abstract

The present invention discloses a self-learning method, system, storage medium and electronic device for automobile assembly, the method comprising: real-time collection of multi-dimensional data in the assembly process; inputting the multi-dimensional data into a distributed cloud computing platform, preprocessing and analyzing the multi-dimensional data using a multi-layer neural network algorithm, and generating quality assessment indicators for assembly steps; storing the processed multi-dimensional data and quality assessment indicators in a distributed database to form a historical data set; training and self-learning the historical data set using a recursive neural network model, and updating the quality prediction model in the assembly process; generating an assembly process optimization plan using the updated quality prediction model to maximize the quality of the next assembly cycle. Using the embodiments of the present invention, a closed loop can be formed through real-time data accumulation, and an assembly process optimization plan can be generated using the updated quality prediction model to adapt to a dynamically changing production environment and improve overall assembly efficiency and quality.

Description

Technical Field

[0001] The present invention belongs to the technical field of automobile assembly, and in particular to an automobile assembly self-learning method, system, storage medium and electronic equipment. Background Art

[0002] In the modern automobile manufacturing industry, with the advancement of technology and the improvement of consumer demand, the accuracy and efficiency of the automobile assembly process are particularly important. Traditional assembly methods often rely on experience and manual adjustments, resulting in quality fluctuations, low efficiency and high rework costs during the production process. These problems not only affect the production efficiency of enterprises, but also reduce the market competitiveness of products.

[0003] With the rapid development of information technology, especially the rise of big data, cloud computing and artificial intelligence technology, the automotive manufacturing industry has ushered in new opportunities for change. Real-time collection of multi-dimensional data during the assembly process, including temperature, pressure, time, worker operation, etc., can provide rich information for the evaluation of assembly quality. However, relying solely on data collection is not enough to solve quality problems in the assembly process. How to efficiently and accurately analyze and use this data to achieve continuous optimization of the assembly process has become a difficult problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a self-learning method, system, storage medium and electronic device for automobile assembly to address the deficiencies in the prior art, form a closed loop through real-time data accumulation, and use an updated quality prediction model to generate an assembly process optimization plan to adapt to the dynamically changing production environment and improve the overall assembly efficiency and quality.

[0005] An embodiment of the present application provides a method for self-learning of automobile assembly, the method comprising:

[0006] During the automobile assembly process, multi-dimensional data of the assembly process is collected in real time;

[0007] Inputting the multi-dimensional data into a distributed cloud computing platform to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps;

[0008] The processed multi-dimensional data and the corresponding quality assessment indicators are stored in a distributed database to form a historical data set;

[0009] Using a time series-based recursive neural network model to train and self-learn historical data sets, the neural network-based quality prediction model used to predict assembly quality in the assembly process is updated to adapt to the dynamic changes in the production environment;

[0010] Using the updated quality prediction model, an assembly process optimization plan is generated to maximize the quality of the next assembly cycle.

[0011] Optionally, the multi-dimensional data is input into a distributed cloud computing platform to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps, including:

[0012] The multi-dimensional data is sharded according to the source and type, each sharded data is allocated by a hash function, and sharded data with similar characteristics are allocated to the same computing node of the distributed cloud computing platform;

[0013] Dynamically allocate computing resources on a distributed cloud computing platform, and initialize a pre-parallel trained multi-layer neural network model for each computing node, wherein the corresponding multi-layer neural network model is allocated according to the shard data, and different multi-layer neural network models are obtained by parallel training of different historical shard data sets;

[0014] On each computing node, the sliced ​​data in the computing node is input into the correspondingly assigned multi-layer neural network model, and the corresponding local quality assessment index is output, which represents the quality assessment of different assembly steps;

[0015] The local quality assessment indicators output by each computing node are weighted averaged to generate a global unified quality assessment indicator.

[0016] Optionally, the use of a recursive neural network model based on a time series to train and self-learn a historical data set to update a quality prediction model based on a neural network in the assembly process for predicting assembly quality includes:

[0017] Standardize and normalize each time series data in the historical data set, and divide the processed time series data into multiple time windows of fixed length, each of which contains multiple time steps;

[0018] Constructing a multi-layer recursive neural network, wherein the multi-layer recursive neural network includes two long short-term memory network layers, one fully connected layer and an output layer, the long short-term memory network layer is used to capture the long-term and short-term dependencies in the time series, and the fully connected layer is used to combine multi-dimensional features to generate the final prediction result;

[0019] The time series data is gradually added to the model according to the time window, and the model parameters are updated through batch training, wherein the latest part of the data is regularly extracted from the historical data set, combined with the current new data, and the model is incrementally updated. The model is updated after each new data input, rather than batch update, to achieve real-time self-learning;

[0020] The RMSprop optimizer is used in combination with adaptive learning rate adjustment. After each training cycle, the learning rate is dynamically adjusted according to the changes in the loss function.

[0021] Optionally, the step of using the updated quality prediction model to generate an assembly process optimization plan to maximize the quality of the next assembly cycle includes:

[0022] Based on the multi-dimensional data in the current assembly process, the updated quality prediction model is used to re-predict new quality assessment indicators;

[0023] According to the new quality evaluation index, the assembly process is optimized using an optimization algorithm to obtain a new assembly process optimization scheme, wherein the objective function of the optimization algorithm is to minimize the frequency of occurrence of low-quality steps and maximize the proportion of high-quality steps, the low-quality steps are assembly steps whose local quality evaluation index is lower than a first preset threshold, and the high-quality steps are assembly steps whose local quality evaluation index is higher than a second preset threshold.

[0024] Another embodiment of the present application provides an automobile assembly self-learning system, the system comprising:

[0025] The acquisition module is used to collect multi-dimensional data of the assembly process in real time during the automobile assembly process;

[0026] A processing module, used for inputting the multi-dimensional data into a distributed cloud computing platform, so as to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm, and generate quality assessment indicators for assembly steps;

[0027] A storage module is used to store the processed multi-dimensional data and the corresponding quality assessment indicators in a distributed database to form a historical data set;

[0028] An update module is used to train and self-learn the historical data set using a recursive neural network model based on a time series, and to update the quality prediction model based on a neural network in the assembly process for predicting assembly quality, so as to adapt to the dynamic changes of the production environment;

[0029] The generation module is used to generate an assembly process optimization plan using the updated quality prediction model to maximize the quality of the next assembly cycle.

[0030] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when running.

[0031] Yet another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described above.

[0032] Compared with the prior art, the present invention provides a self-learning method for automobile assembly. During the automobile assembly process, multi-dimensional data of the assembly process is collected in real time; the multi-dimensional data is input into a distributed cloud computing platform, and the multi-dimensional data is preprocessed and analyzed by a multi-layer neural network algorithm to generate quality evaluation indicators of the assembly steps; the processed multi-dimensional data and the corresponding quality evaluation indicators are stored in a distributed database to form a historical data set; a recursive neural network model based on a time series is used to train and self-learn the historical data set, and a quality prediction model based on a neural network for predicting assembly quality in the assembly process is updated; an assembly process optimization plan is generated by using the updated quality prediction model to maximize the quality of the next assembly cycle, so that a closed loop can be formed through real-time data accumulation, and an assembly process optimization plan is generated by using the updated quality prediction model to adapt to the dynamically changing production environment and improve the overall assembly efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A hardware structure block diagram of a computer terminal for a self-learning method for automobile assembly provided by an embodiment of the present invention;

[0034] Figure 2 A schematic diagram of a process flow of a self-learning method for automobile assembly provided by an embodiment of the present invention;

[0035] Figure 3 A schematic diagram of the structure of an automobile assembly self-learning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but should not be construed as limiting the present invention.

[0037] The embodiment of the present invention first provides a method for self-learning in automobile assembly, which can be applied to electronic devices such as computer terminals, specifically ordinary computers, etc.

[0038] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal for a self-learning method for automobile assembly provided by an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the computer terminal may also include a transmission device 106 for communication functions and an input and output device 108. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0039] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / modules corresponding to the automobile assembly self-learning method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0040] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0041] See also Figure 2 , an embodiment of the present invention provides a method for self-learning of automobile assembly, which may include the following steps:

[0042] S201, during the automobile assembly process, real-time collection of multi-dimensional data of the assembly process;

[0043] In the modern automobile manufacturing industry, the quality of the assembly process is directly related to the performance and safety of the final product. Therefore, it is crucial to collect multi-dimensional data in the assembly process in real time. These data include not only key process parameters (such as temperature, pressure, vibration, etc.), but also production environment factors (such as humidity, light, noise, etc.) and operator feedback (such as worker operation time, work efficiency and error records, etc.). Through real-time monitoring and collection of these multi-dimensional data, manufacturing companies can promptly identify potential quality problems, optimize production processes, and improve overall assembly efficiency and product quality. The collection of multi-dimensional data can provide a rich source of information for subsequent analysis and model building, allowing companies to use data-driven decisions to dynamically adjust and optimize assembly processes, thereby improving production flexibility and responsiveness.

[0044] Specifically, various types of sensors can be installed at key locations on the assembly line, including temperature sensors, pressure sensors, acceleration sensors, photoelectric sensors, etc. These sensors can collect environmental and process data related to the assembly process in real time. Through the Internet of Things (IoT) technology, the data obtained by each sensor is transmitted to the central processing unit through a wireless network. After the data arrives at the center, data cleaning and preprocessing are performed, including noise removal, missing value filling and data standardization to ensure data accuracy and reliability. A streaming data processing framework (such as Apache Kafka) is used to divert and transmit real-time data to different storage areas. For data that requires high-frequency analysis (such as temperature, pressure, etc.), it is stored in an in-memory database for fast reading; while historical data is stored in a long-term storage relational database or distributed database for subsequent analysis and model training.

[0045] S202, inputting the multi-dimensional data into a distributed cloud computing platform, so as to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps;

[0046] The step of inputting multi-dimensional data into the distributed cloud computing platform is mainly to solve the problem that traditional computing resources are insufficient to process complex assembly data. Through the cloud computing platform, enterprises can obtain almost unlimited computing resources to support large-scale data processing and the training of deep learning models. The multi-layer neural network algorithm can capture the complex nonlinear relationships in the data, which is particularly important for quality assessment in the assembly process. The key to this process lies in the preprocessing and analysis of the data. The preprocessing step can remove noise, fill in missing values, and standardize the data to ensure the quality of the input data. At the same time, through the analysis of the multi-layer neural network algorithm, the key features that affect the assembly quality can be extracted, and finally the quality assessment indicators of the assembly steps can be generated. These indicators provide real-time quality feedback to production managers, so that problems in the assembly process can be identified and corrected in a timely manner, thereby improving overall production efficiency and product quality.

[0047] Specifically, the multi-dimensional data may be sharded according to source and type, and each sharded data may be allocated through a hash function, so that sharded data with similar characteristics may be allocated to the same computing node of the distributed cloud computing platform;

[0048] During the assembly process, data from different sensors (such as temperature sensors, pressure sensors, accelerometers, etc.) are collected in real time, and the source identifier of each data is recorded. For each piece of data, information such as timestamp and sensor type needs to be attached for subsequent processing. The sharding rules are defined according to the source (such as sensor type, location) and type (such as process parameters, environmental parameters, etc.) of the data. For example, the data of each sensor can be grouped according to the time series, each group of data has the same time interval (such as one group every 10 seconds), and the data is divided into shards of a specified size (such as 100 data per shard). The source identifier of each data is hashed using a hash function (such as MD5 or SHA256) to generate a unique hash value to ensure uniform distribution of data. The hash value provides a digital space, and the data is distributed according to the hash value, so that data with similar sources and types can be distributed to the same computing node. Each computing node in the distributed cloud computing platform predefines the data processing capacity and resource configuration. Through the hash value, each data shard is pointed to a specific computing node. The purpose of this is to ensure data processing efficiency and effectively utilize the resources of computing nodes. For example, data from multiple sensors can be processed in parallel on the same node, reducing network latency.

[0049] Dynamically allocate computing resources on a distributed cloud computing platform, and initialize a pre-parallel trained multi-layer neural network model for each computing node, wherein the corresponding multi-layer neural network model is allocated according to the shard data, and different multi-layer neural network models are obtained by parallel training of different historical shard data sets;

[0050] In the cloud computing platform, the resource usage of each computing node is monitored in real time, including CPU utilization, memory utilization, and storage load. By monitoring data, the load capacity of each node is dynamically evaluated to allocate computing resources as needed. For each computing node, a multi-layer neural network model is initialized, and the architecture of the model can be optimized based on the previous analysis results. For example, a model architecture suitable for specific data features such as a convolutional neural network (CNN) or a long short-term memory network (LSTM) can be selected. Ensure that each model has enough layers and nodes to meet the complexity of data processing. According to the allocated data shards, different shard data are transmitted to the corresponding computing node. On this node, the model is trained in parallel using historical shard data to ensure that the model of each node uses its own shard data set, thereby improving the training effect and the adaptability of the model. During the training process, ensure that the model of each node can report the training results regularly, and synchronize the model parameters back to the central node after the training is completed. In this way, a centralized model update mechanism can be established. Although each node is trained independently, it can eventually be integrated to form a globally optimal model to provide support for subsequent data analysis.

[0051] On each computing node, the sliced ​​data in the computing node is input into the correspondingly assigned multi-layer neural network model, and the corresponding local quality assessment index is output, which represents the quality assessment of different assembly steps;

[0052] Format the raw data from the shards into an input format acceptable to the model. This includes converting the data into the form of numerical matrices and vectors, and performing necessary standardization and normalization to maintain data consistency. Input the formatted data into the corresponding multi-layer neural network model to perform forward propagation calculations. The model analyzes the data through its hidden layers and output layers to generate output results, specifically local quality assessment indicators for each assembly step. After processing the shard data, each computing node stores the generated local quality assessment indicators (such as predicted values, evaluation scores, etc.) in the local database and aggregates them to the central database of the cloud platform for subsequent global evaluation and analysis.

[0053] The local quality assessment indicators output by each computing node are weighted averaged to generate a global unified quality assessment indicator.

[0054] The local quality assessment indicators outputted by each computing node can be collected regularly through the central coordination module of the distributed cloud computing platform. The output of each node should include the local quality assessment value of the corresponding assembly step and its associated assembly step number or identifier. A weighting coefficient is assigned to each computing node based on factors such as the total amount of historical data processed by each computing node, the processing capacity, and the reliability of the indicator. The weighting coefficient can be calculated by the following formula:

[0055] Among them, W_i is the weight coefficient of node i, N_i is the number of historical processed data of node i, N_j is the number of historical processed data of node j, and k is the total number of nodes.

[0056] The local quality evaluation index of each computing node is weighted. Specifically, for the local quality evaluation index Q_i from node i, its weighted value Q_{weighted,i} can be calculated by the following formula:

[0057] The weighted indicators of all computing nodes are accumulated to form the value of global quality assessment. The weighted quality assessment values ​​of all nodes are summarized to generate a globally unified quality assessment indicator Q_{global}. The calculation formula is:

[0058] To facilitate understanding, the global evaluation indicator Q_{global} can be mapped to a predetermined quality rating standard (such as a grade system, a percentage system, etc.) to output clear quality evaluation results for reference by management or production line operators.

[0059] The above steps ensure that by taking the weighted average of the local quality evaluation indicators of each computing node, a global evaluation indicator that comprehensively reflects the quality of the entire assembly process is formed, thereby improving the level of intelligent management of the production process.

[0060] S203, storing the processed multi-dimensional data and the corresponding quality assessment indicators in a distributed database to form a historical data set;

[0061] In the self-learning method of automobile assembly, this process not only ensures the persistence and accessibility of data, but also provides rich and reliable basic data for subsequent analysis and model updates. As an accumulation of experience, historical data sets can reflect the changing trends and potential problems in the production process, and provide support for the optimization and quality assurance of the manufacturing process. The formed historical data set helps enterprises accumulate production experience in the long term, and then optimize the production process; by analyzing historical data, potential defect patterns and key factors affecting quality in the assembly process can be identified; rich historical data provides more samples for the training of machine learning models, so that the model can more accurately capture the complex relationships in the assembly process; over time, the accumulation of historical data can help the system continuously improve and adapt to changes in the production environment, thereby reducing quality fluctuations.

[0062] Specifically, before storing the processed multi-dimensional data and quality assessment indicators, it is necessary to first remove redundant and erroneous data through the data cleaning module to ensure the high quality of the incoming data. For the collected data and the generated quality assessment indicators, a dynamic decision tree algorithm is used to automatically generate labels for each data record based on the contextual information of the current assembly (such as time, node status, etc.). These labels include timestamp, assembly step identification, equipment status, quality level and other information. Different types of data are stored in multiple distributed database nodes. By selecting a specific hash function (such as MD5 or SHA-256), data is mapped to different storage areas according to the feature mixing amount, so that similar data can be read and managed more efficiently.

[0063] Among them, in order to optimize storage space and reading efficiency, time series data compression algorithms (such as FastPFor or Zstandard) are used to compress and store time series data in historical data sets. A fast access path for time series data is established through indexes to ensure high performance when reading. In addition, in database design, a version control mechanism can be implemented, that is, each time new data is stored, a new version number is generated, and the historical data set can retain the old version to support time backtracking and data analysis. Linked lists or tree structures are used to manage versions so that data changes at different time points can be traced and analyzed. The processed data and evaluation indicators on any node are sent to the main database through a message queue system (such as Kafka) to ensure real-time and high availability of data storage without interfering with the main data processing process. Data integrity verification is added during data synchronization to ensure data consistency during transmission.

[0064] Through the above steps, the processed multi-dimensional data and quality assessment indicators can be efficiently and reliably stored in a distributed database, forming a rich historical data set, providing a strong basis for subsequent quality prediction models and assembly process optimization.

[0065] S204, using a recursive neural network model based on a time series to train and self-learn the historical data set, and update the quality prediction model based on the neural network in the assembly process for predicting the assembly quality, so as to adapt to the dynamic changes of the production environment;

[0066] The historical data set is trained and self-learned using a time series-based recurrent neural network (RNN) model. By inputting the time series of historical data, the network can identify the dependencies and change patterns between different time points in the assembly process, thereby updating the quality prediction model. The output of the model will be a prediction of future assembly quality, which is adaptively adjusted based on real-time input data and historical information.

[0067] Through time series analysis, RNN can capture dynamic changes in the production environment, adjust quality predictions in a timely manner, and enhance the accuracy and reliability of predictions; as the model's self-learning process continues, decision makers can obtain more accurate quality assessment tools, thereby effectively controlling the production process and reducing the risk of unqualified products; RNN can process and memorize long-term and short-term dependencies in time series, enhance the robustness of the model in complex production environments, and adapt to changing production conditions; accumulated historical data continuously provides new training samples for the model, driving the continuous optimization of the quality prediction model over time and forming a closed-loop feedback mechanism.

[0068] In an embodiment of the present invention, the quality prediction model used is a model based on a recurrent neural network (RNN), in particular a long short-term memory (LSTM) network. The model is suitable for processing time series data and can capture dynamic changes and long-term and short-term dependencies related to the assembly process.

[0069] 1. Long Short-Term Memory Network (LSTM): As a variant of RNN, LSTM can retain important information in sequence data and avoid the gradient vanishing problem in traditional RNN by introducing the mechanism of forget gate, input gate and output gate. This enables LSTM to better handle long sequence data, such as continuously collected assembly process data. Specifically, LSTM can effectively capture the dependencies between different time points in the assembly process, thereby obtaining accurate predictions of future assembly quality.

[0070] 2. Input and output: The input of the model is pre-processed multi-dimensional data, including assembly environment, process parameters, etc., which are standardized and normalized to form structured time series data. The output of the model is the quality assessment indicators for the assembly steps, including the prediction of future quality status. These indicators provide a basis for subsequent production decisions.

[0071] The updated quality prediction model can generate quality assessment indicators based on multi-dimensional data collected in real time, thereby providing production managers with timely feedback on the quality of the assembly process. This process can identify potential problems in real time and reduce the risk of substandard products; the model's self-learning ability enables it to continuously adapt as the production environment changes, improving the accuracy and reliability of predictions. By continuously accumulating historical data, the model can capture the complex nonlinear relationships in the assembly process, thereby making flexible quality predictions under dynamic conditions; with the help of the generated quality assessment indicators, companies can optimize assembly processes more scientifically and improve overall production efficiency. This not only reduces low-quality steps in production, but also effectively increases the proportion of high-quality steps, providing support for the sustainable development of the company.

[0072] Specifically, each time series data in the historical data set can be standardized and normalized, and the processed time series data can be divided into multiple time windows of fixed length, each time window containing multiple time steps;

[0073] This step aims to preprocess historical data to eliminate the dimensional effects between different features and simplify the complexity of the data. Standardization and normalization can ensure the stability of the model during training and prevent certain features from affecting model learning due to excessive numerical ranges. The purpose of dividing the time window is to form an input format suitable for time series models (such as LSTM). Standardization and normalization enhance the comparability of data and reduce the convergence time and risk during training. The division of the time window enables the model to obtain sufficient contextual information, thereby improving the accuracy and effect of the prediction.

[0074] The required multi-dimensional data, such as temperature, humidity, equipment operating status, etc., can be extracted from the historical data set and organized into a time series. Standardization is performed on each feature, the mean and standard deviation of each feature are calculated, and each data point is transformed using the Z-score formula. At the same time, Min-Max normalization is applied to scale the data to the range of [0,1] to ensure that all features are at the same magnitude. The length of the time window can be determined (such as 10 time steps), and then the time series data can be divided into multiple windows of fixed length using the sliding window method. Each window contains data of several time steps, providing rich contextual information for the model. For example, if each time window includes data from the past 5 minutes and data is collected once a minute, each window contains information from 5 time steps.

[0075] Constructing a multi-layer recursive neural network, wherein the multi-layer recursive neural network includes two long short-term memory network layers, one fully connected layer and an output layer, the long short-term memory network layer is used to capture the long-term and short-term dependencies in the time series, and the fully connected layer is used to combine multi-dimensional features to generate the final prediction result;

[0076] In this step, a recursive neural network architecture is designed to effectively capture the time-dependent characteristics in the data. RNN can remember important information from the past through its special structure (such as LSTM units), so as to take into account the historical context when making predictions. Such a network architecture can effectively process time series data, and by capturing long-term and short-term dependencies, it can improve the performance of the model when processing complex time series, thereby improving the accuracy of quality predictions.

[0077] You can choose LSTM as the basic unit to build a multi-layer RNN. Design two LSTM layers, each with 128 LSTM units, to enhance the expressiveness of the model. Add a fully connected layer (Dense layer) after the LSTM layer to transform the LSTM output into a high-level feature representation. Set the number of neurons in this layer (such as 64) and activate it using the ReLU activation function. Add a linear output layer after the fully connected layer. The output layer will generate prediction results based on the output of the previous layer. Usually, the number of neurons in this layer is equal to the prediction target (such as the number of quality assessment indicators).

[0078] The time series data is gradually added to the model according to the time window, and the model parameters are updated through batch training, wherein the latest part of the data is regularly extracted from the historical data set, combined with the current new data, and the model is incrementally updated. The model is updated after each new data input, rather than batch update, to achieve real-time self-learning;

[0079] This step involves gradually feeding the processed time series data into the network for training. By training batch by batch, data can be effectively utilized and the model can be updated at any time, enabling it to learn and adapt to new information in real time. Real-time self-learning capabilities ensure that the model is constantly updated to adapt to changes in the production environment. This means that the model is no longer dependent on a fixed data set, but can instantly reflect new production conditions, improving the timeliness and accuracy of predictions.

[0080] The data segments of each time window can be input into the model batch by batch, and 5-10 windows can be selected for each batch of data. Ensure that the model can effectively process and learn new data patterns in each iteration. Set up a mechanism so that when new data is generated (for example, new assembly data records), the system automatically integrates the new data into the latest historical data set and retains the most recent M window data for training. At this time, the sliding window technology is used to update the model parameters only with the most recent data instead of retraining the entire model. After each new data is input, the model parameters are updated using the corresponding optimization algorithm (such as RMSprop) without the need to batch process all historical data. This incremental update method ensures that the model can promptly reflect changes that occur in the actual assembly process.

[0081] The RMSprop optimizer is used in combination with adaptive learning rate adjustment. After each training cycle, the learning rate is dynamically adjusted according to the changes in the loss function.

[0082] In this step, the RMSprop optimizer is selected and the adaptive learning rate is used to improve the training efficiency and stability. By monitoring the loss function and dynamically adjusting the learning rate, the model can be converged more accurately. The adaptive learning rate mechanism can reduce the training instability caused by the fixed learning rate, and can adjust the parameters more carefully during the model convergence stage, which helps to improve the model performance and prediction accuracy.

[0083] You can select RMSprop as the optimizer in model training, and set the initial learning rate to 0.001 to ensure that the learning speed is appropriate at the beginning of the optimization process. After each training, calculate the current loss function and compare it with the previous loss. If the loss decreases, the learning rate remains unchanged; if the loss does not change significantly, adjust the current learning rate by multiplying it by a decay factor (such as 0.95) to ensure that the training process gradually stabilizes.

[0084] S205, using the updated quality prediction model, generating an assembly process optimization plan to maximize the quality of the next assembly cycle.

[0085] This step involves using the updated quality prediction model to analyze the current assembly process. New quality evaluation indicators are generated by inputting multi-dimensional data collected in real time into the prediction model. These evaluation indicators will be used to evaluate each step in the assembly process and provide a basis for subsequent optimization decisions.

[0086] The core purpose of this step is to identify potential quality problems and opportunities in the assembly process through scientific predictive analysis. With the updated predictive model, you can quickly respond to the dynamically changing production environment, adjust the assembly process, and develop an optimization plan. Through the continuously updated quality prediction model, the current assembly status can be reflected in real time to ensure that the optimization decision of the assembly process is based on the latest data and predictive analysis; the generated optimization plan helps to reduce the occurrence of low-quality steps in the assembly process, improve the overall product quality, and reduce the cost of rework and maintenance in the later stage; by optimizing the assembly steps, the process efficiency in production can be improved, and unnecessary links can be reduced, thereby speeding up production and saving manpower and material resources.

[0087] Step S202 focuses on using a multi-layer neural network (such as a fully connected or convolutional network in deep learning) to comprehensively analyze the multi-dimensional data collected in real time. The focus is on extracting features from the original data and identifying the key factors affecting assembly quality. The generated quality assessment indicators are preliminary judgments on each assembly step, usually in some form of scoring or classification, reflecting the overall quality status of the step.

[0088] The core of this step S205 is to use the updated quality prediction model (i.e., the LSTM-based model) for real-time prediction. This process is based on the preliminary indicators generated in step S202, combined with real-time assembly data for further analysis and prediction, focusing on the dynamics of time series. The newly generated quality evaluation indicators not only reflect the quality of the current assembly step, but also are real-time predictions based on historical data and dynamic changes in time series. This enables it to more accurately reflect the actual situation of the assembly process and provide a more forward-looking quality control method.

[0089] In general, step S202 focuses on feature extraction and preliminary analysis of data, laying the foundation for subsequent model training and optimization; while step S205 uses real-time data to conduct more dynamic and accurate quality assessment based on the existing model. Through this step-by-step and layered processing method, the entire system can more effectively adapt to complex production environments, provide real-time feedback and optimize the assembly process, and form a dynamic closed-loop quality management mechanism.

[0090] Specifically, the updated quality prediction model can be used to re-predict new quality assessment indicators based on the multi-dimensional data in the current assembly process;

[0091] Collect multi-dimensional data (such as temperature, pressure, assembly speed, etc.) generated during the assembly process in real time and input these data into the updated quality prediction model. The model uses the input data provided to calculate the latest quality assessment indicators to reflect the current production status.

[0092] Real-time prediction enables production managers to understand the quality status of the current assembly process in a timely manner, providing a basis for subsequent decision-making; by predicting new quality indicators, managers can adjust assembly strategies and processes with greater confidence and reduce human errors; timely quality assessment can help quickly identify potential problems, reduce the production of unqualified products, and ensure that products meet quality standards.

[0093] According to the new quality evaluation index, the assembly process is optimized using an optimization algorithm to obtain a new assembly process optimization scheme, wherein the objective function of the optimization algorithm is to minimize the frequency of occurrence of low-quality steps and maximize the proportion of high-quality steps, the low-quality steps are assembly steps whose local quality evaluation index is lower than a first preset threshold, and the high-quality steps are assembly steps whose local quality evaluation index is higher than a second preset threshold.

[0094] Using new quality assessment indicators, we apply optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) to analyze and optimize the assembly process, develop more effective assembly plans, and improve the operating steps in the production process. By defining the goal of the optimization algorithm, that is, by adjusting the assembly process, the frequency of low-quality steps is as low as possible, while the proportion of high-quality steps is increased to the optimal level. To this end, two thresholds are set to determine which steps are of low quality and high quality, respectively.

[0095] First, analyze the new quality assessment indicators to clarify which steps are considered low-quality steps (e.g., steps where the local quality assessment indicator is lower than the first preset threshold) and which steps are considered high-quality steps (e.g., steps where the local quality assessment indicator is higher than the second preset threshold). Set specific thresholds, such as 70% and below as the low-quality judgment standard and 70% and above as the high-quality judgment standard.

[0096] Define the objective function of the optimization algorithm, which aims to minimize the frequency of low-quality steps and maximize the proportion of high-quality steps at the same time. The objective function can be formalized as:

[0097] in, and It is a weight coefficient set according to the company's production strategy and quality requirements, responsible for balancing the emphasis on optimizing low-quality and high-quality steps.

[0098] Select appropriate optimization algorithms according to the characteristics of the optimization target, such as genetic algorithms, simulated annealing algorithms, or particle swarm optimization. Based on the characteristics of different algorithms, design appropriate implementation frameworks to ensure that input data and objective functions can be effectively processed. Input the current assembly process information and new quality assessment indicators into the optimization algorithm. The algorithm will simulate according to the defined objective function and explore the arrangement and execution order of different assembly steps to achieve the optimization goal. During the optimization process, the algorithm will continuously evaluate the quality of each step to provide real-time feedback and adjustments. After the algorithm obtains a new assembly process optimization solution, perform feasibility verification. By simulating the assembly process, verify the effectiveness and quality control effect of the optimized process solution in actual operation. You can choose to conduct pilot tests in small batch production to evaluate the quality changes of the optimized steps. Based on the results of the pilot test, provide feedback on the operation effect of the optimization algorithm. Evaluate the specific proportion of low-quality and high-quality steps, and adjust the parameters of the objective function and the optimization strategy according to the actual situation, so as to further improve the quality of the assembly process in the next round of optimization.

[0099] Through the above implementation method, combined with new quality assessment indicators and optimization goals, the optimization effect of the assembly process can be effectively improved, the incidence of low-quality steps can be reduced, and the proportion of high-quality steps can be increased, thereby ensuring the continuous improvement of the overall assembly quality.

[0100] It can be seen that in the automobile assembly process, multi-dimensional data of the assembly process is collected in real time; the multi-dimensional data is input into the distributed cloud computing platform, and the multi-dimensional data is preprocessed and analyzed by using a multi-layer neural network algorithm to generate quality assessment indicators for the assembly steps; the processed multi-dimensional data and the corresponding quality assessment indicators are stored in a distributed database to form a historical data set; the historical data set is trained and self-learned by using a time series-based recursive neural network model, and the quality prediction model based on the neural network in the assembly process for predicting the assembly quality is updated; the updated quality prediction model is used to generate an assembly process optimization plan to maximize the quality of the next assembly cycle, so that a closed loop can be formed through real-time data accumulation, and the updated quality prediction model is used to generate an assembly process optimization plan to adapt to the dynamically changing production environment and improve the overall assembly efficiency and quality.

[0101] Another embodiment of the present invention provides a vehicle assembly self-learning system, see Figure 3 , the system may include:

[0102] The acquisition module 301 is used to collect multi-dimensional data of the assembly process in real time during the automobile assembly process;

[0103] A processing module 302 is used to input the multi-dimensional data into a distributed cloud computing platform to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps;

[0104] The storage module 303 is used to store the processed multi-dimensional data and the corresponding quality assessment indicators in a distributed database to form a historical data set;

[0105] An updating module 304 is used to train and self-learn the historical data set using a recursive neural network model based on a time series, and update the quality prediction model based on a neural network in the assembly process for predicting assembly quality, so as to adapt to the dynamic changes of the production environment;

[0106] The generation module 305 is used to generate an assembly process optimization plan using the updated quality prediction model to maximize the quality of the next assembly cycle.

[0107] It can be seen that in the automobile assembly process, multi-dimensional data of the assembly process is collected in real time; the multi-dimensional data is input into the distributed cloud computing platform, and the multi-dimensional data is preprocessed and analyzed by using a multi-layer neural network algorithm to generate quality assessment indicators for the assembly steps; the processed multi-dimensional data and the corresponding quality assessment indicators are stored in a distributed database to form a historical data set; the historical data set is trained and self-learned by using a time series-based recursive neural network model, and the quality prediction model based on the neural network in the assembly process for predicting the assembly quality is updated; the updated quality prediction model is used to generate an assembly process optimization plan to maximize the quality of the next assembly cycle, so that a closed loop can be formed through real-time data accumulation, and the updated quality prediction model is used to generate an assembly process optimization plan to adapt to the dynamically changing production environment and improve the overall assembly efficiency and quality.

[0108] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0109] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps:

[0110] S201, during the automobile assembly process, real-time collection of multi-dimensional data of the assembly process;

[0111] S202, inputting the multi-dimensional data into a distributed cloud computing platform, so as to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps;

[0112] S203, storing the processed multi-dimensional data and the corresponding quality assessment indicators in a distributed database to form a historical data set;

[0113] S204, using a recursive neural network model based on a time series to train and self-learn the historical data set, and update the quality prediction model based on the neural network in the assembly process for predicting the assembly quality, so as to adapt to the dynamic changes of the production environment;

[0114] S205, using the updated quality prediction model, generating an assembly process optimization plan to maximize the quality of the next assembly cycle.

[0115] It can be seen that in the automobile assembly process, multi-dimensional data of the assembly process is collected in real time; the multi-dimensional data is input into the distributed cloud computing platform, and the multi-dimensional data is preprocessed and analyzed by using a multi-layer neural network algorithm to generate quality assessment indicators for the assembly steps; the processed multi-dimensional data and the corresponding quality assessment indicators are stored in a distributed database to form a historical data set; the historical data set is trained and self-learned by using a time series-based recursive neural network model, and the quality prediction model based on the neural network in the assembly process for predicting the assembly quality is updated; the updated quality prediction model is used to generate an assembly process optimization plan to maximize the quality of the next assembly cycle, so that a closed loop can be formed through real-time data accumulation, and the updated quality prediction model is used to generate an assembly process optimization plan to adapt to the dynamically changing production environment and improve the overall assembly efficiency and quality.

[0116] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0117] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0118] Specifically, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0119] S201, during the automobile assembly process, real-time collection of multi-dimensional data of the assembly process;

[0120] S202, inputting the multi-dimensional data into a distributed cloud computing platform, so as to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps;

[0121] S203, storing the processed multi-dimensional data and the corresponding quality assessment indicators in a distributed database to form a historical data set;

[0122] S204, using a recursive neural network model based on a time series to train and self-learn the historical data set, and update the quality prediction model based on the neural network in the assembly process for predicting the assembly quality, so as to adapt to the dynamic changes of the production environment;

[0123] S205, using the updated quality prediction model, generating an assembly process optimization plan to maximize the quality of the next assembly cycle.

[0124] It can be seen that in the automobile assembly process, multi-dimensional data of the assembly process is collected in real time; the multi-dimensional data is input into the distributed cloud computing platform, and the multi-dimensional data is preprocessed and analyzed by using a multi-layer neural network algorithm to generate quality assessment indicators for the assembly steps; the processed multi-dimensional data and the corresponding quality assessment indicators are stored in a distributed database to form a historical data set; the historical data set is trained and self-learned by using a time series-based recursive neural network model, and the quality prediction model based on the neural network in the assembly process for predicting the assembly quality is updated; the updated quality prediction model is used to generate an assembly process optimization plan to maximize the quality of the next assembly cycle, so that a closed loop can be formed through real-time data accumulation, and the updated quality prediction model is used to generate an assembly process optimization plan to adapt to the dynamically changing production environment and improve the overall assembly efficiency and quality.

[0125] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. A self-learning method for automobile assembly, characterized in that: The method comprises: During the automobile assembly process, multi-dimensional data of the assembly process is collected in real time; Inputting the multi-dimensional data into a distributed cloud computing platform to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps; The processed multi-dimensional data and the corresponding quality assessment indicators are stored in a distributed database to form a historical data set; Using a time series-based recursive neural network model, the historical data set is trained and self-learned to update the quality prediction model based on the neural network used to predict the assembly quality during the assembly process to adapt to the dynamic changes in the production environment; Using the updated quality prediction model, an assembly process optimization plan is generated to maximize the quality of the next assembly cycle; The multi-dimensional data is input into a distributed cloud computing platform to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps, including: The multi-dimensional data is sharded according to the source and type, each sharded data is allocated by a hash function, and sharded data with similar characteristics are allocated to the same computing node of the distributed cloud computing platform; Dynamically allocate computing resources on a distributed cloud computing platform, and initialize a pre-parallel trained multi-layer neural network model for each computing node, wherein the corresponding multi-layer neural network model is allocated according to the shard data, and different multi-layer neural network models are obtained by parallel training of different historical shard data sets; On each computing node, the sliced ​​data in the computing node is input into the correspondingly assigned multi-layer neural network model, and the corresponding local quality assessment index is output, which represents the quality assessment of different assembly steps; The local quality assessment indicators output by each computing node are weighted averaged to generate a global unified quality assessment indicator.

2. The method according to claim 1, characterized in that The method uses a recursive neural network model based on a time series to train and self-learn a historical data set, and updates a quality prediction model based on a neural network in the assembly process for predicting assembly quality, including: Standardize and normalize each time series data in the historical data set, and divide the processed time series data into multiple time windows of fixed length, each of which contains multiple time steps; Constructing a multi-layer recursive neural network, wherein the multi-layer recursive neural network includes two long short-term memory network layers, one fully connected layer and an output layer, the long short-term memory network layer is used to capture the long-term and short-term dependencies in the time series, and the fully connected layer is used to combine multi-dimensional features to generate the final prediction result; The time series data is gradually added to the model according to the time window, and the model parameters are updated through batch training, wherein the latest part of the data is regularly extracted from the historical data set, combined with the current new data, and the model is incrementally updated. The model is updated after each new data input, rather than batch update, to achieve real-time self-learning; The RMSprop optimizer is used in combination with adaptive learning rate adjustment. After each training cycle, the learning rate is dynamically adjusted according to the changes in the loss function.

3. The method according to claim 2, characterized in that The method of using the updated quality prediction model to generate an assembly process optimization plan to maximize the quality of the next assembly cycle includes: The multi-dimensional data of the current assembly process uses the updated quality prediction model to re-predict new quality assessment indicators; According to the new quality evaluation index, the assembly process is optimized using an optimization algorithm to obtain a new assembly process optimization scheme, wherein the objective function of the optimization algorithm is to minimize the frequency of occurrence of low-quality steps and maximize the proportion of high-quality steps, the low-quality steps are assembly steps whose local quality evaluation index is lower than a first preset threshold, and the high-quality steps are assembly steps whose local quality evaluation index is higher than a second preset threshold.

4. An automobile assembly self-learning system, characterized in that: The system comprises: The acquisition module is used to collect multi-dimensional data of the assembly process in real time during the automobile assembly process; A processing module, used for inputting the multi-dimensional data into a distributed cloud computing platform, so as to pre-process and analyze the multi-dimensional data using a multi-layer neural network algorithm to generate quality assessment indicators for assembly steps; A storage module is used to store the processed multi-dimensional data and the corresponding quality assessment indicators in a distributed database to form a historical data set; An update module is used to train and self-learn the historical data set using a recursive neural network model based on a time series, and to update the quality prediction model based on a neural network in the assembly process for predicting assembly quality, so as to adapt to the dynamic changes of the production environment; A generation module, for generating an assembly process optimization plan using the updated quality prediction model to maximize the quality of the next assembly cycle; The processing module is specifically used for: The multi-dimensional data is sharded according to the source and type, each sharded data is allocated by a hash function, and sharded data with similar characteristics are allocated to the same computing node of the distributed cloud computing platform; Dynamically allocate computing resources on a distributed cloud computing platform, and initialize a pre-parallel trained multi-layer neural network model for each computing node, wherein the corresponding multi-layer neural network model is allocated according to the shard data, and different multi-layer neural network models are obtained by parallel training of different historical shard data sets; On each computing node, the sliced ​​data in the computing node is input into the correspondingly assigned multi-layer neural network model, and the corresponding local quality assessment index is output, which represents the quality assessment of different assembly steps; The local quality assessment indicators output by each computing node are weighted averaged to generate a global unified quality assessment indicator.

5. The system according to claim 4, characterized in that The update module is specifically used for: Standardize and normalize each time series data in the historical data set, and divide the processed time series data into multiple time windows of fixed length, each of which contains multiple time steps; Constructing a multi-layer recursive neural network, wherein the multi-layer recursive neural network includes two long short-term memory network layers, one fully connected layer and an output layer, the long short-term memory network layer is used to capture the long-term and short-term dependencies in the time series, and the fully connected layer is used to combine multi-dimensional features to generate the final prediction result; The time series data is gradually added to the model according to the time window, and the model parameters are updated through batch training, wherein the latest part of the data is regularly extracted from the historical data set, combined with the current new data, and the model is incrementally updated. The model is updated after each new data input, rather than batch update, to achieve real-time self-learning; The RMSprop optimizer is used in combination with adaptive learning rate adjustment. After each training cycle, the learning rate is dynamically adjusted according to the changes in the loss function.

6. The system according to claim 5, characterized in that The generation module is specifically used for: Based on the multi-dimensional data in the current assembly process, the updated quality prediction model is used to re-predict new quality assessment indicators; According to the new quality evaluation index, the assembly process is optimized using an optimization algorithm to obtain a new assembly process optimization scheme, wherein the objective function of the optimization algorithm is to minimize the frequency of occurrence of low-quality steps and maximize the proportion of high-quality steps, the low-quality steps are assembly steps whose local quality evaluation index is lower than a first preset threshold, and the high-quality steps are assembly steps whose local quality evaluation index is higher than a second preset threshold.

7. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 3 when executed.

8. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 3.

Images