Search

CN-121980956-A - Mold life assessment method based on machine learning

CN121980956ACN 121980956 ACN121980956 ACN 121980956ACN-121980956-A

Abstract

The invention relates to the technical field of machine learning, and discloses a mold life assessment method based on machine learning, which comprises the steps of collecting mold data and extracting characteristics related to a physical failure mode of a mold; the method comprises the steps of performing self-adaptive decomposition, adjusting decomposition parameters to obtain signal components, constructing an integrated model to obtain a predicted value of the residual service life of the die, analyzing the characteristic contribution degree, optimizing the input characteristic set of the model, outputting maintenance decision suggestions based on analysis results, deploying the integrated model to edge computing equipment, and adjusting evaluation strategies according to degradation characteristics of the die in different life cycles to realize real-time service life evaluation and decision support. The invention can obviously improve the accuracy and reliability of the prediction of the service life of the die, enhance the transparency of the model and realize the closed-loop management from the prediction to the maintenance decision.

Inventors

  • CHEN YONGBIN

Assignees

  • 南通汉冶金属制品有限公司

Dates

Publication Date
20260505
Application Date
20251017

Claims (6)

  1. 1. A machine learning based mold life assessment method, comprising the steps of: step S1, collecting mold manufacturing process parameters, working condition monitoring signals and workshop environment data, and extracting features related to a physical failure mode of a mold from the online working condition monitoring signals; Step S2, carrying out self-adaptive decomposition on the collected working condition monitoring signals, and adjusting decomposition parameters to obtain signal components by taking the consistency of a decomposition result and a theoretical physical model as a target; s3, constructing an integrated model for fusion of deep learning and traditional machine learning, and carrying out fusion processing on the characteristics and the signal components to obtain a predicted value of the residual service life of the die; S4, analyzing the feature contribution degree based on the interpretable artificial intelligence technology, inputting a feature set according to a contribution degree optimization model, and outputting maintenance decision advice based on an analysis result; Step S5, deploying the integrated model to edge computing equipment, and adjusting an evaluation strategy according to degradation characteristics of the die in different life cycles to realize real-time life evaluation and decision support; wherein in step S2, the method further comprises the following sub-steps: s2-1, initializing parameters of a fully integrated empirical mode decomposition self-adaptive noise algorithm, and setting an initial range of a noise standard deviation, an initial range of a total average frequency and a maximum iteration frequency; S2-2, inputting the vibration signal into a fully integrated empirical mode decomposition adaptive noise algorithm to obtain a plurality of groups of eigenmode functions and a residual sequence; S2-3, calculating a correlation coefficient of a decomposition residual sequence and a theoretical abrasion loss calculated based on an Alchard abrasion model, and calculating sample entropy of an intrinsic mode function component, wherein the specific formula of the Alchard abrasion model is as follows: wherein dW is the wear depth, K is the wear factor, The normal pressure, u is the slip speed, and H is the die hardness; s2-4, adjusting parameters of a fully integrated empirical mode decomposition adaptive noise algorithm by a grid search method until a correlation coefficient reaches a preset threshold value and sample entropy is minimum, and determining decomposition parameters; s2-5, removing eigenvalue function components irrelevant to die degradation according to the correlation of the components and physical characteristics, and reserving effective eigenvalue function components and residual sequences, wherein the independence of the die degradation means that the correlation of the components is lower than a preset threshold; wherein in step S3, the method further comprises the following sub-steps: S3-1, constructing a multi-layer architecture of an encoder for processing long-term trend characteristics, a cyclic neural network for processing short-term trend characteristics, an attention fusion layer and a meta-decision layer, wherein the attention fusion layer comprises a spatial attention sub-layer and a temporal attention sub-layer, the spatial attention sub-layer protrudes out of the cavity stress concentration region characteristics, and the temporal attention sub-layer strengthens failure early-stage characteristics; S3-2, combining the extracted characteristics, the screened signal components, manufacturing process parameters and workshop environment data to form a characteristic vector, taking the residual working cycle number as a sample label, and dividing a training set and a verification set according to a preset proportion; s3-3, adopting a composite loss function fusing a prediction error and a physical consistency penalty term, and selecting an adaptive optimization algorithm for training, wherein the training strategy comprises an adaptive learning rate adjustment mechanism and an early stopping mechanism; S3-4, inputting a training set into the integrated model for iterative training, dynamically adjusting model super-parameters, and evaluating performance by using a verification set until the model is converged and then storing a final model, wherein the model super-parameters comprise the number of hidden layer nodes of the converter and the random inactivation probability of a gating circulation unit; The parameters of the Alchard abrasion model in the step S2-3 are determined by collecting a plurality of groups of standard abrasion test data for different die materials, and obtaining a special abrasion factor of the materials by adopting least square fitting; the attention fusion layer in step S3-1 calculates the dynamic weight according to the following formula: , wherein, And For the attention score, T is the current time, T is the total time step of the input sequence, k is the cyclic index, all time steps from 1 to T are traversed, For the attention weight at time t, The method comprises the steps of outputting feature vectors for a preamble network, v and R are learnable parameter matrixes, b is a bias vector, generating a space weight graph by a space attention sub-layer through convolution operation, giving higher weight to the features of a stress concentration area, and giving higher weight to the features of the early failure stage by a time attention sub-layer.
  2. 2. The machine learning based mold life assessment method of claim 1, wherein: wherein in step S1, the method further comprises the following sub-steps: S1-1, acquiring manufacturing process parameters from a manufacturing execution system, wherein the manufacturing process parameters comprise mold material parameters, heat treatment process parameters and machining precision parameters; S1-2, collecting working condition monitoring signals through a multi-source sensor network, wherein the working condition monitoring signals comprise vibration signals, acoustic emission signals and temperature signals; S1-3, collecting workshop environment data through a workshop environment sensor, wherein the workshop environment data comprises environment temperature, humidity and dust concentration; S1-4, calculating kurtosis index and root mean square ratio from the vibration acceleration signal, and using the kurtosis index and root mean square ratio as early warning characteristics of the transition of abrasion from a stable period to an acceleration period; S1-5, counting the energy of an emergency and the cumulative quantity of ringing counts from an acoustic emission signal stream, and taking the cumulative quantity as a judging basis for microscopic crack initiation and propagation; S1-6, calculating the maximum temperature rise rate and the thermal cycle asymmetry from the thermocouple temperature measurement data, and taking the maximum temperature rise rate and the thermal cycle asymmetry as characteristic parameters for evaluating the accumulation of thermal fatigue damage.
  3. 3. The method for evaluating the service life of a mold based on machine learning according to claim 2, wherein the step S1 further comprises frequency domain feature extraction, namely performing Fourier transform on vibration signals to extract specific frequency band energy duty ratio corresponding to macroscopic abrasion and microscopic crack, performing wavelet packet decomposition on acoustic emission signals to extract specific frequency band kurtosis factors corresponding to crack propagation as sensitive features of crack propagation.
  4. 4. The machine learning based mold life assessment method of claim 1, wherein: wherein in step S4, the method further comprises the following sub-steps: S4-1, inputting a verification set into a model with training completed, calculating contribution degree of each input characteristic to a residual service life predicted value by adopting an interpretive artificial intelligence technology, generating a characteristic contribution degree time-varying curve, and determining contribution degree duty ratios of each key characteristic of different life periods, wherein the different life periods comprise a new model stage, a middle stage and a failure early stage; S4-2, after the prediction of the residual service life of the preset times is completed, ranking the contribution degrees of the features, reserving the features of the preset quantity, eliminating the redundant features, and updating the input feature set of the model; S4-3, establishing a high contribution characteristic and a maintenance action preset knowledge base, and matching and outputting corresponding maintenance decision suggestions from the preset knowledge base according to the type and the numerical value of the high contribution characteristic.
  5. 5. The machine learning based mold life assessment method of claim 4, wherein the interpretable artificial intelligence technique in step S4-1 is a saprolidine additive interpretation algorithm modified by physical prior constraints, the modification comprising assigning high computational weight to features directly related to mold failure mechanisms and low computational weight to indirect influencing features, the feature contribution time-varying curve being dynamically updated using a sliding time window mechanism.
  6. 6. The machine learning based mold life assessment method of claim 1, wherein: Wherein in step S5, the method further comprises the following sub-steps: S5-1, deploying the trained integrated model, a dynamic feature selection function and a maintenance decision mapping function which are realized based on the interpretability analysis to edge computing equipment together, wherein the edge computing equipment supports industrial Ethernet communication; s5-2, in a new model stage, calling manufacturing process data to calibrate an initial life baseline, reducing new model evaluation errors, in a middle stage, strengthening trend analysis of working condition monitoring signals, capturing wear accumulation characteristics, and in a failure early stage, starting a high-frequency sampling mode, and enhancing capturing capacity of microscopic crack characteristics; S5-3, the edge computing equipment receives the monitoring data in real time, automatically executes the processes of feature extraction, signal decomposition, model reasoning and decision output, and outputs the predicted value of the residual service life and maintenance advice; S5-4, the operation data accumulated regularly are used for adjusting the model deployed on the edge computing equipment, the bottom layer network parameters are fixed during adjustment, and the meta-decision layer parameters are optimized.

Description

Mold life assessment method based on machine learning The application relates to a divisional application of an application with the application number 202511487920.5 and the application name of 'a method for evaluating the service life of a die based on machine learning', which is filed on the 10 th and 17 th of 2025. Technical Field The invention relates to the technical field of machine learning, in particular to a mold life assessment method based on machine learning. Background The service life of the die is directly influenced on the production efficiency and the product quality as key process equipment in industrial production. In actual production, the mold often degrades gradually due to various physical failure modes including wear, fatigue, crack propagation, ultimately leading to failure. Traditional die life assessment methods mostly depend on periodic maintenance, experience judgment or prediction based on a physical model, and the methods often have the problems of strong subjectivity, low prediction precision and incapability of real-time response. With the development of intelligent manufacturing and industrial Internet of things technology, the real-time acquisition of the running state data of the die through the sensor is possible. However, due to the complex working conditions of the die, heterogeneous data, and various failure mechanisms, how to extract effective features from massive monitoring data and build an accurate life prediction model is still a great challenge. The existing method often fails to fully integrate manufacturing process parameters, environment data and real-time monitoring signals, and lacks a self-adaptive decomposition and feature screening mechanism for multi-mode data, so that the model has insufficient generalization capability and poor interpretation. In addition, most of the existing life prediction models only stay in a cloud centralized processing stage, and the requirements of production sites on instantaneity and reliability are difficult to meet. Although the edge calculation has the advantages of low delay and high response, how to lighten and adapt a complex machine learning model to edge equipment and realize dynamic evaluation strategy adjustment at different stages of life cycle is still a technical problem to be solved. Therefore, the invention provides a mold life assessment method based on machine learning. Disclosure of Invention The invention aims to solve the problems of low prediction precision, poor real-time performance and weak model interpretability of the die life in the prior art, and provides a die life assessment method based on machine learning. In order to achieve the above purpose, the invention adopts the following technical scheme that the die life assessment method based on machine learning comprises the following steps: step S1, collecting mold manufacturing process parameters, working condition monitoring signals and workshop environment data, and extracting features related to a physical failure mode of a mold from the online working condition monitoring signals; Step S2, carrying out self-adaptive decomposition on the collected working condition monitoring signals, and adjusting decomposition parameters to obtain signal components by taking the consistency of a decomposition result and a theoretical physical model as a target; s3, constructing an integrated model for fusion of deep learning and traditional machine learning, and carrying out fusion processing on the characteristics and the signal components to obtain a predicted value of the residual service life of the die; S4, analyzing the feature contribution degree based on the interpretable artificial intelligence technology, inputting a feature set according to a contribution degree optimization model, and outputting maintenance decision advice based on an analysis result; and S5, deploying the integrated model to edge computing equipment, and adjusting an evaluation strategy according to degradation characteristics of the die in different life cycles to realize real-time life evaluation and decision support. Further, in step S1, the method further includes the following sub-steps: S1-1, acquiring manufacturing process parameters from a manufacturing execution system, wherein the manufacturing process parameters comprise mold material parameters, heat treatment process parameters and machining precision parameters; S1-2, collecting working condition monitoring signals through a multi-source sensor network, wherein the working condition monitoring signals comprise vibration signals, acoustic emission signals and temperature signals; S1-3, collecting workshop environment data through a workshop environment sensor, wherein the workshop environment data comprises environment temperature, humidity and dust concentration; S1-4, calculating kurtosis index and root mean square ratio from the vibration acceleration signal, and using the kurtosis index and root mean square ra