CN-121638694-B - Big data-based automobile aluminum ornament management method

CN121638694BCN 121638694 BCN121638694 BCN 121638694BCN-121638694-B

Abstract

The invention discloses a big data-based automobile aluminum ornament management method, which belongs to the technical field of aluminum ornament management and specifically comprises the steps of collecting raw material purity, an extrusion process and environment temperature and humidity data through a sensor, transmitting the raw material purity, the extrusion process and the environment temperature and humidity data to a cloud end, carrying out structural association storage according to unique product serial numbers, and carrying out time sequence alignment, denoising and normalization processing on the data by utilizing a dynamic time alignment algorithm to construct a full life cycle feature matrix. The method comprises the steps of carrying out iterative training on a feature matrix based on a long-short-term memory network, establishing a quality dynamic prediction model, calculating quality scores of aluminum ornaments in real time by the model in actual production, automatically judging and marking abnormal products and defect types lower than a threshold value, finally analyzing the model by applying a layer correlation propagation algorithm, positioning key feature dimensions causing the abnormality, and generating a closed-loop correction instruction aiming at adjustable parameters to realize intelligent optimization of a production process.

Inventors

ZHENG XINBIN
Rov Temerich
CHEN YONG

Assignees

安德佳（福建）铝饰科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (6)

1. The automobile aluminum ornament management method based on big data is characterized by comprising the following steps of: s1, acquiring raw material purity, extrusion process parameters and environmental temperature and humidity data of an aluminum ornament through sensors deployed in various production links, transmitting the data to a cloud server, and carrying out structural association storage according to unique product serial numbers; s2, performing time sequence alignment on the data stored in the structural association by utilizing a dynamic time warping algorithm, eliminating noise interference and performing normalization processing to construct an aluminum ornament full life cycle feature matrix containing raw materials, processes and environmental features; S3, dividing the full life cycle feature matrix of the aluminum ornament into a training set and a verification set, inputting a long-term and short-term memory network for iterative training, and establishing an aluminum ornament quality dynamic prediction model taking the production process features as input and the finished product quality index as output; s4, collecting current production process data in real time, inputting the current production process data into an aluminum ornament quality dynamic prediction model, calculating the quality score of the current aluminum ornament, automatically judging the aluminum ornament to be an abnormal product and marking a specific defect type if the score is lower than a standard threshold value; s5, aiming at the aluminum ornament judged to be an abnormal product, analyzing the activation state of neurons in the model by adopting a layer correlation propagation algorithm, positioning key feature dimensions causing quality abnormality, and if the feature dimensions belong to adjustable process parameters, generating a closed loop correction instruction aiming at the parameters, wherein the specific process is as follows: inputting the characteristic matrix of the aluminum ornament determined to be abnormal into an aluminum ornament quality dynamic prediction model, locking neuron nodes indicating abnormal categories in an output layer, and setting the activation value of the neuron nodes as an initial correlation total value; reversely distributing the initial correlation total value according to the network weight by using a layer class correlation propagation rule, and gradually and reversely pushing the initial correlation total value through a full-connection layer and an implicit layer to calculate the correlation score of each characteristic dimension in an input layer; comparing the relevance scores of all feature dimensions of the input layer, and screening out the feature dimension with the largest value as a key feature dimension; Judging whether the key feature dimension belongs to a preset adjustable process parameter list; If yes, searching a preset process compensation strategy matched with the key feature dimension, calculating a physical adjustment value required by the parameter, and generating a closed-loop correction instruction containing the target equipment ID and the adjustment value; the specific way of calculating the relevance score of each feature dimension in the input layer is as follows: calculating the weighted product of the neuron activation value of the full-connection layer and the connection weight, defining the proportion of the weighted product of the full-connection layer to the sum of all weighted products of the layer as a contribution coefficient, and distributing the initial correlation total value of the output layer to each neuron of the full-connection layer according to the contribution coefficient; Taking the correlation value obtained by the neurons of the full-connection layer as reverse propagation input, traversing the hidden layer network structure aiming at the time expansion structure of the long-period memory network, calculating the excitation intensity of the neurons of the previous layer to the neurons of the next layer according to the interlayer connection weight and the activation state of the neurons, and transmitting the correlation value forwards layer by layer according to the excitation intensity proportion; when the correlation value is reversely propagated to the input layer, the correlation components transmitted by all nodes of the first hidden layer to the corresponding nodes of the input layer are collected, summation operation is carried out to obtain the accumulated correlation value of the nodes, and the accumulated correlation value is established as the correlation score of each characteristic dimension of the input layer.
2. The method for managing automotive aluminum trim parts based on big data according to claim 1, wherein in the step S1, the process of transmitting to the cloud server and performing structured association storage according to the unique product serial number is as follows: the cloud server receives an encapsulation data packet uploaded by the edge gateway, wherein the data packet comprises raw material purity, extrusion process parameters and environment temperature and humidity values which are bound with unique product serial numbers; The server analyzes the data packet to extract a unique product serial number as a main key of the database, and constructs a mapping index table containing raw material purity fields, extrusion process parameter fields and environment temperature and humidity fields; And writing the analyzed data values into storage field positions corresponding to the unique product serial numbers in the mapping index table to form structured associated data entries with the unique product serial numbers as indexes.
3. The method for managing automotive aluminum trim parts based on big data according to claim 1, wherein in the step S2, the process of constructing the full life cycle feature matrix of the aluminum trim parts including raw materials, process and environmental features is as follows: Reading a data sequence in the structured associative memory, selecting a standard process time sequence as a reference, and applying a dynamic time warping algorithm to map the data sequence to a unified time axis; performing Kalman filtering on the data sequences on the unified time axis, removing high-frequency random noise, and generating a smooth data sequence with reserved trend characteristics; the smooth data sequence is traversed to calculate an extremum, and the numerical value is mapped to a closed interval from zero to one by using a range transformation formula to finish data normalization processing; And splicing the normalized raw material purity, the extrusion process parameters and the environmental temperature and humidity data according to time dimension, and constructing the full life cycle characteristic matrix of the aluminum ornament.
4. The method for managing automotive aluminum trim parts based on big data according to claim 3, wherein the specific way of performing kalman filtering on the data sequences on the unified time axis to reject high-frequency random noise and generate the smooth data sequences with the trend feature is as follows: Constructing a linear state space model aiming at a data sequence on a unified time axis, setting a system state transition matrix and an observation matrix, and calculating a priori state estimation value and a priori error covariance matrix at the current moment by using a posterior state estimation value at the previous moment; calculating Kalman gain by combining the prior error covariance matrix and the observation noise covariance matrix, and carrying out weighted correction on residual errors between the observation value and the prior state estimation value at the current moment in the data sequence by using the Kalman gain to calculate a posterior state estimation value at the current moment; Updating the posterior error covariance matrix for iterative computation at the next moment, extracting posterior state estimated values obtained by computation at each moment on a unified time axis, and forming a smooth data sequence with reserved trend characteristics.
5. The method for managing automotive aluminum trim parts based on big data according to claim 1, wherein in the step S3, the process of creating the dynamic prediction model of aluminum trim part quality using the production process feature as input and the product quality index as output is as follows: dividing the full life cycle feature matrix of the aluminum ornament into a training set matrix and a verification set matrix which are not overlapped with each other according to the time sequence; Constructing a long-term and short-term memory network structure comprising an input layer, an hidden layer and a fully connected output layer, and initializing network weight parameters and bias items; Inputting the training set matrix into a long-short-term memory network to execute forward propagation, and calculating the deviation between a predicted result and a finished product quality index by using a loss function; and updating the network weight according to the deviation gradient by adopting a back propagation algorithm until the verification set error converges to output an aluminum ornament quality dynamic prediction model.
6. The method for managing automotive aluminum trim parts based on big data according to claim 1, wherein in the step S4, the process of collecting the current production process data in real time, inputting the data into the dynamic prediction model of aluminum trim parts quality, calculating the quality score of the current aluminum trim parts, determining abnormal products and marking specific defect types is as follows: Reading the sensor value in the current production process in real time, and multiplexing preprocessing rules in the training stage to execute time sequence alignment and normalization to generate a current feature matrix to be detected; Inputting the current feature matrix to be detected into an aluminum ornament quality dynamic prediction model, and calculating by a model output layer to obtain the quality score and defect type probability distribution of the current aluminum ornament; Comparing the quality score with a preset standard threshold value, and if the quality score is smaller than the preset standard threshold value, generating an abnormal product judgment result and activating a defect recognition logic; And matching the maximum probability value index in the defect type probability distribution with a pre-stored defect type database, and associating and marking a specific defect type name for the abnormal product.

Description

Big data-based automobile aluminum ornament management method Technical Field The invention relates to the technical field of aluminum ornament management, in particular to an automobile aluminum ornament management method based on big data. Background The rapid development of the automobile industry has improved requirements of consumers on the appearance texture and the reliability of parts of the automobile, and the automobile aluminum ornament has the advantages of light weight, corrosion resistance, attractive appearance and the like, and is widely applied to the assembly of automobile interiors and exteriors. In the process of production management of automobile aluminum ornaments, sensor technology is gradually introduced in the industry to collect data of production links, which covers key procedures such as raw material inspection, extrusion molding, surface treatment and the like, and relates to recording and storing basic data such as raw material purity detection data, extrusion process parameters (such as extrusion temperature, extrusion speed and pressure maintaining time), production environment temperature and humidity and the like. Meanwhile, the existing management system mostly comprises a basic quality inspection link, whether the aluminum ornament is qualified or not is judged through the modes of finished product appearance detection, size measurement and the like, a basic management flow of data acquisition, storage and finished product inspection is formed, and preliminary data and flow support is provided for quality control of the aluminum ornament. However, the existing automobile aluminum ornament management system lacks large data-driven quality association analysis capability, quality data of all links are stored in an isolated form, and a quantitative association model between raw material purity, processing technological parameters, environment temperature and humidity and aluminum ornament finished product quality cannot be established. When quality problems such as surface defects and size deviation appear in the aluminum ornament, the problem sources cannot be rapidly located through data tracing, and only manual item-by-item investigation can be relied on, so that the labor cost of quality control is increased, the solution period of the quality problem is prolonged, the timeliness of the quality control is reduced, and strict standards of the automobile part industry on quality tracing are difficult to meet. Disclosure of Invention The invention aims to provide a big data-based automobile aluminum ornament management method, which solves the problems in the background technology: the aim of the invention can be achieved by the following technical scheme: a management method of an automobile aluminum ornament based on big data comprises the following steps: s1, acquiring raw material purity, extrusion process parameters and environmental temperature and humidity data of an aluminum ornament through sensors deployed in various production links, transmitting the data to a cloud server, and carrying out structural association storage according to unique product serial numbers; s2, performing time sequence alignment on the data stored in the structural association by utilizing a dynamic time warping algorithm, eliminating noise interference and performing normalization processing to construct an aluminum ornament full life cycle feature matrix containing raw materials, processes and environmental features; S3, dividing the full life cycle feature matrix of the aluminum ornament into a training set and a verification set, inputting a long-term and short-term memory network for iterative training, and establishing an aluminum ornament quality dynamic prediction model taking the production process features as input and the finished product quality index as output; s4, collecting current production process data in real time, inputting the current production process data into an aluminum ornament quality dynamic prediction model, calculating the quality score of the current aluminum ornament, automatically judging the aluminum ornament to be an abnormal product and marking a specific defect type if the score is lower than a standard threshold value; S5, aiming at the aluminum ornament judged to be abnormal, analyzing the activation state of neurons in the model by adopting a layer correlation propagation algorithm, positioning key feature dimensions causing quality abnormality, and if the feature dimensions belong to adjustable process parameters, generating a closed loop correction instruction aiming at the parameters. In the step S1, the process of transmitting to the cloud server and carrying out structured association storage according to the unique product serial number is as follows: the cloud server receives an encapsulation data packet uploaded by the edge gateway, wherein the data packet comprises raw material purity, extrusion process parameters and environment temperatu