Mold life prediction and maintenance strategy making method based on big data analysis

Abstract

The invention discloses a mold life prediction and maintenance strategy making method based on big data analysis, and particularly relates to the field of mold life prediction and maintenance, and the method comprises the following steps: collecting and associating multi-source heterogeneous data in a mold production process; comprising process parameter time sequence data, on-line monitoring time sequence data and structured and image data of off-line inspection; performing time domain alignment and collaborative feature extraction on the data, and constructing a structured training data set taking a production cycle as a unit; inputting the data set into a dynamic evolution prediction model, outputting a mold health state score, and generating a dynamic maintenance strategy through a multi-objective optimization model in combination with production context information; and finally, visually displaying the strategy, converting the strategy into a control instruction, issuing the control instruction to a production management system, and automatically triggering maintenance execution. The method can realize high-precision life prediction, dynamically optimize the maintenance decision, significantly improve the service life and production efficiency of the mold, and reduce the maintenance cost and production risk.

Description

Technical Field

[0001] This invention relates to the field of mold life prediction and maintenance technology, and more specifically, to a method for formulating mold life prediction and maintenance strategies based on big data analysis. Background Technology

[0002] In modern manufacturing processes such as injection molding and die casting, molds are core equipment for ensuring product quality and production efficiency. Their health directly affects the continuity of the production line and the product qualification rate. Currently, the industry relies heavily on manual experience, fixed-cycle maintenance, or single-type sensor monitoring for mold life prediction and maintenance. However, these methods still have some drawbacks in actual use. For example, existing technologies rely on periodic shutdowns for disassembly and inspection, which is not only inefficient and costly, but also cannot monitor the internal wear and crack propagation of the mold in real time, easily leading to inaccurate predictions or untimely maintenance.

[0003] On the other hand, single sensor data has limited dimensions, making it difficult to fully capture the multimodal characteristics of the mold degradation process. The predictive model has weak generalization ability, especially poor adaptability to new molds or process changes. Furthermore, existing maintenance strategies are mostly static rules, which cannot be dynamically adjusted according to the actual production status (such as order urgency, inventory status, and resource availability), often resulting in production interruptions or resource waste. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for predicting mold life and formulating maintenance strategies based on big data analysis. The method addresses the problems mentioned in the background art, such as inaccurate predictions and untimely maintenance due to reliance on manual experience and regular inspections, weak generalization ability of prediction models caused by analysis of a single data source, and resource waste and production interruptions caused by static maintenance strategies being difficult to adapt to dynamic production contexts.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting mold life and formulating maintenance strategies based on big data analysis, comprising: S1: data acquisition and fusion: acquiring and associating multi-source heterogeneous data in the mold production process, wherein the multi-source heterogeneous data includes time-series data of process parameters from production equipment, time-series data of online monitoring from sensors, and structured and image data of offline inspection; S2: Collaborative Feature Engineering: Process the fused multi-source heterogeneous data to generate a structured training dataset with the production cycle as the basic unit; S3: Intelligent Prediction and Decision Making: The structured training dataset is input into a dynamic evolution prediction model to obtain the health status score of the mold's remaining useful life; the health status score, together with the current production context information, is input into the adaptive decision engine to calculate and generate a dynamic maintenance strategy. S4: Strategy Output and Execution: Output dynamic maintenance strategies to user terminals for visualization, and convert dynamic maintenance strategies into control commands and send them to the production management system to automatically trigger corresponding maintenance execution actions.

[0006] Preferably, the multi-source heterogeneous data in the associated mold production process are associated and fused using the mold ID and production cycle ID as unique identifiers; The process parameter timing data includes injection pressure, injection speed, and mold temperature timing data. The online monitoring time-series data includes online monitoring time-series data from vibration sensors and acoustic emission sensors; The structured and image data includes wear images of key parts of the mold and structured dimensional measurement data from the machine vision system.

[0007] Preferably, the collaborative feature engineering process involves processing the fused multi-source heterogeneous data, including temporal alignment and collaborative feature extraction of the multi-source heterogeneous data.

[0008] Preferably, the time-domain alignment uses a single production cycle as the basic time unit, and maps online monitoring time-series data and process parameter time-series data with different sampling frequencies to each production cycle in a unified manner through timestamp alignment and data interpolation.

[0009] Preferably, the collaborative feature extraction extracts time-domain statistical features and frequency-domain spectral features from online monitoring time-series data, extracts visual features representing surface texture, crack length and area from image data, and fuses the extracted multivariate features with the process parameter features within the corresponding production cycle to construct a cycle-feature matrix as a structured training dataset.

[0010] Preferably, the dynamic evolution prediction model is constructed through an ensemble learning algorithm, and transfer learning technology is used to transfer the feature extraction layer of the trained model to the prediction model of the new model.

[0011] Preferably, the dynamic evolution prediction model is equipped with a model update triggering mechanism. When the prediction error continuously exceeds the threshold or the production materials are changed, the model parameters are automatically started to incrementally learn using the latest production data.

[0012] Preferably, the adaptive decision engine is implemented through a multi-objective optimization model. The optimization objective function of this model simultaneously considers production costs, product quality risks, equipment safety risks, as well as production context information such as the urgency of current production orders, spare parts inventory status, and available maintenance personnel information. By solving this function, a comprehensive cost dynamic maintenance strategy is obtained.

[0013] Preferably, the dynamic maintenance strategy includes immediate preventative replacement, delaying execution to the next planned maintenance window, and degraded use by adjusting subsequent production process parameters to extend mold life.

[0014] Preferably, the conversion of dynamic maintenance strategies into control instructions involves automatically generating and issuing a corresponding standardized set of control instructions to the production management system based on the specific type of the dynamic maintenance strategy; immediately executing preventive replacements corresponds to generating preventive maintenance work order creation, dispatch, and resource reservation instructions; and adjusting subsequent production process parameters corresponds to generating production equipment process parameter adjustment instructions.

[0015] The technical effects and advantages of this invention are as follows: 1. This invention constructs a high-dimensional feature set that comprehensively characterizes the degradation state of molds by fusing multi-source heterogeneous data and performing collaborative feature engineering, overcoming the limitations of traditional single data source analysis and significantly improving the accuracy and reliability of mold remaining life prediction. 2. This invention adopts a dynamic evolutionary prediction model that combines ensemble learning and transfer learning, and introduces an incremental learning mechanism to enable the model to have self-evolution capabilities, adapt to scenarios such as new molds and process changes, significantly reduce the amount of training data required for the target mold, and improve the generalization and practicality of the model. 3. This invention utilizes a multi-objective optimization adaptive decision engine to comprehensively consider production costs, quality risks, safety risks, and real-time production context information. It generates a dynamic cost maintenance strategy and can automatically convert it into control commands to drive the execution system. This achieves closed-loop automation from prediction to execution, effectively avoiding production interruptions, optimizing the allocation of maintenance resources, and significantly improving production efficiency and economic benefits. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the collaborative feature engineering process of the present invention; Figure 3 This is a schematic diagram of the intelligent prediction and decision-making process of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] As attached Figure 1-3 The method for predicting mold life and formulating maintenance strategies based on big data analysis, as shown, includes: S1: Data Acquisition and Fusion: Collect and correlate multi-source heterogeneous data in the mold production process. The multi-source heterogeneous data includes time-series data of process parameters from production equipment, time-series data of online monitoring from sensors, and structured and image data of offline inspection.

[0019] It should be specifically noted that the process parameter timing data includes injection pressure, injection speed, and mold temperature timing data; The online monitoring time-series data includes online monitoring time-series data from vibration sensors and acoustic emission sensors; The structured and image data includes wear images of key parts of the mold and structured dimensional measurement data from the machine vision system.

[0020] It should be further noted that the process parameter timing data uses PT124G series pressure sensors, LVT-200 series speed sensors, and PT100 / PT1000 temperature sensors. The pressure sensors are installed at the nozzle of the barrel and the sprue gate, with the core parameter being the injection pressure P, a sampling frequency of 1-5Hz, an accuracy of ±0.5%FS, and a range of 0-300MPa. The speed sensors are installed at the end of the screw drive motor shaft and the mold closing mechanism, with the core parameters being the injection speed v and the mold closing speed vc. clamp The sampling frequency is 1-5Hz, the accuracy is ±0.1mm / s, and the measurement range is 0-600mm / s. Injection speed affects melt erosion intensity, and mold closing speed is related to collision risk. Temperature sensors are installed on the cavity sidewalls, gate, and barrel areas. The core parameters are mold temperature T and barrel temperature T. barrel Sampling frequency 1Hz, accuracy ±0.1℃, measurement range 0-400℃.

[0021] The online monitoring time-series data uses VS-300 series vibration sensors and AE-500 series acoustic emission sensors. The vibration sensors are installed symmetrically on the moving and fixed mold bases, with the core parameter being vibration acceleration *a*, a sampling frequency of 500-2000Hz, an accuracy of ±0.01g, and a range of ±50g, capturing vibration signals from mold opening and closing impacts and wear. The acoustic emission sensors are installed in the center of the fixed mold base or near the cavity, with the core parameters being the acoustic emission signal amplitude *A* and the count *N*. count The sampling frequency is 1000-5000Hz, the accuracy is ±1dB, and the range is 0-120dB. It uses acoustic energy to determine the generation and propagation of cracks.

[0022] Structured and image data are obtained using MD series digital micrometers / ten-thousand-digit calipers and MV-CA series industrial machine vision systems. The digital calipers are used for manual or automated measurement stations, with the core parameter being the critical cavity dimension D. The sampling frequency is once every 5-20 cycles, with an accuracy of ±0.001mm and a measuring range of 0-50mm, directly quantifying dimensional wear. The machine vision system is installed directly above the cavity after mold opening, equipped with a ring / strip light source. The core parameter is the cavity surface image I, with a sampling frequency of once every 1-5 cycles, a resolution of not less than 2048*1536, and a frame rate of not less than 5fps, detecting defects such as surface wear, scratches, and cracks.

[0023] It should be specifically noted that the multi-source heterogeneous data in the associated mold production process are associated and fused using the mold ID and production cycle ID as unique identifiers.

[0024] It should be further explained that the mold ID adopts the format of "material type-mold application code-serial number", such as TP-IM-005, where TP represents thermoplastic plastic, IM represents injection mold, and 005 is the serial number of the same type of mold; the production cycle ID adopts the format of "production equipment number-date-cycle number", such as IM-08-20241001-0123, where IM-08 is the injection molding machine number, 20241001 is the production date, and 0123 is the 123rd cycle of the equipment on that day.

[0025] Further explanation is needed regarding the specific data fusion execution process, which includes: connecting real-time sensor data to the EC-1000 series edge computing gateway via an industrial Ethernet network supporting Profinet and EtherNet / IP protocols with a transmission latency of ≤100ms; directly connecting image data to the gateway via Gigabit Ethernet, and connecting offline dimensional data via a manual input terminal + barcode scanning associated cycle ID or an automated measurement equipment API; using the 3σ criterion to eliminate outliers caused by sensor faults and signal interference; filling short-term missing data (≤3 sampling points) with linear interpolation, and filling long-term missing data (>3 sampling points) with the average of adjacent cycles of the same process; finally, establishing a mold-cycle-data association table in the database, using the mold ID + production cycle ID as the joint primary key, and associating process parameters, online monitoring data, and offline data of the same cycle to form a complete single-cycle dataset.

[0026] It should be further explained that when the same mold is temporarily replaced by a different device, an additional device ID is added as a joint identifier on the basis of mold ID + production cycle ID, forming a three-dimensional association key of mold ID - device ID - production cycle ID; when the same device switches molds, a mold change timestamp field is added. If the device switches molds within the same production cycle, the cycle is split according to the mold change timestamp, and the corresponding mold data is associated with each cycle.

[0027] S2: Collaborative Feature Engineering: Process the fused multi-source heterogeneous data to generate a structured training dataset with the production cycle as the basic unit.

[0028] It should be specifically noted that the collaborative feature engineering process involves processing the fused multi-source heterogeneous data, including temporal alignment and collaborative feature extraction of the multi-source heterogeneous data.

[0029] It should be specifically noted that the time-domain alignment uses a single production cycle as the basic time unit, and maps online monitoring time-series data and process parameter time-series data with different sampling frequencies to each production cycle in a unified manner through timestamp alignment and data interpolation.

[0030] It should be further noted that time-domain alignment is based on a single production cycle T. cycle The basic time unit is 20-120 seconds in length, set according to the mold process. All multi-source sensor data are synchronized and uniformly sampled. First, the time window is calibrated, and the start time of mold closing for each cycle is recorded by the injection molding machine PLC signal. start Until the end of mold opening t end Time window [t] start , t end As a benchmark for data alignment, it is synchronized with the production stage; based on this, frequency unification is performed. For online monitoring parameters acquired at high frequencies, a linear interpolation algorithm is used to map them to the same sampling time points as the process parameters, with a frequency of 1-5Hz, thereby generating N equally spaced time points throughout the entire cycle. The interpolation formula is: ,in, The target value obtained after interpolation calculation represents the value at time point [time]. The vibration acceleration value at that location, and It is the high-frequency data that is closest to the target time point. The two timestamps before and after, and For the corresponding measured values, offline data with a set of results only after each cycle is directly associated with the time window of its corresponding cycle; finally, each production cycle is regularized into N consecutive time slices, each time slice contains a set of synchronized process parameters and online monitoring parameters, and is associated with a set of offline test data as a whole to form a regularized multidimensional time series sample.

[0031] It should be specifically noted that the collaborative feature extraction extracts time-domain statistical features and frequency-domain spectral features from online monitoring time-series data, extracts visual features representing surface texture, crack length and area from image data, and fuses the extracted multivariate features with the process parameter features within the corresponding production cycle to construct a cycle-feature matrix as a structured training dataset.

[0032] It should be further explained that collaborative feature extraction covers four major categories of general features: time domain, frequency domain, visual, and process and structured features, as detailed below: Time-domain statistical feature extraction is applied to vibration signal a, acoustic emission signal A, and counting signal N. count Six characteristics are calculated within each process cycle, including: mean. This is used to reflect the average strength of the signal, where N is the total number of signal sampling points in one process cycle, and x... i Let i be a discrete sequence of signal sample values, where i = 1, 2, ..., N; and variance. Used to characterize the degree of signal fluctuation; peak value Used to detect transient impact components; peak-to-peak value Used to describe the range of signal fluctuations; kurtosis Used to determine the existence of an impact event; skewness It is used to evaluate the symmetry of signal distribution.

[0033] Frequency domain spectral feature extraction is performed by applying a Fast Fourier Transform to the vibration signal a to obtain the spectral amplitude A(f), with a frequency range of 0 to... fs is the sampling frequency, and the peak frequency is extracted. Failure type is associated with characteristic frequency offset; spectral centroid Used to reflect the dominant frequency position of energy concentration; spectral bandwidth , used to characterize the dispersion of the frequency distribution, where and The frequencies corresponding to 90% and 10% of the accumulated energy.

[0034] Visual feature extraction is achieved by sequentially converting the acquired image I to grayscale ( The image processing steps include: R, G, and B representing the red, green, and blue channel components of the image pixels, Gaussian filtering (5x5 convolution, σ = 0.8 to 1.2), Canny edge detection (low threshold 50-80, high threshold 150-200), and then contrast calculation based on a grayscale matrix with pixel spacing d = 1 to 3 and angles of 0°, 45°, 90°, and 135°. , used to characterize the degree of wear, where P(i,j) is the gray-level co-occurrence matrix; correlation Used to reflect texture regularity; energy It is used to evaluate surface smoothness; and to quantify crack geometry and length through contour tracing and connected region analysis. , where N p Here, k represents the number of pixels in the crack, k is the pixel equivalent, and area... , where N area This represents the total number of pixels in the cracked area.

[0035] Process and structured feature extraction involves extracting two types of non-sensor features: process parameter features, including mean injection pressure μ. P Maximum injection speed v max Average mold temperature (μ) T and average mold closing speed μ vclamp Structural features include the measured cavity dimension D and dimensional deviations. ,in For design dimensions, when If the dimensions exceed the tolerance range, it is considered to be out of tolerance.

[0036] It should be further explained that the period-feature matrix is ​​constructed by arranging time-domain, frequency-domain, visual, process, and structured features according to the production cycle row to form an m*n structured training dataset X, where m ≥ 1000 periods and n = 35-40 features.

[0037] S3: Intelligent Prediction and Decision Making: The structured training dataset is input into a dynamic evolution prediction model to obtain the health status score of the mold's remaining useful life; the health status score, together with the current production context information, is input into the adaptive decision engine to calculate and generate a dynamic maintenance strategy.

[0038] It should be further explained that the dynamic evolution prediction model is built on an ensemble learning + transfer learning + incremental learning framework, and is used to predict the health status score H of the mold based on the input feature X, ranging from 0 to 100, where 0 represents complete failure and 100 represents a completely new state; its implementation scheme is as follows: The basic prediction model uses the XGBoost algorithm, integrating 80 to 120 decision trees. Core parameters are dynamically set based on the mold complexity: maximum tree depth. The learning rate η is 0.01-0.05, and the subsample ratio is 0.7-0.9. The goal of model training is to minimize the mean squared error loss function. ,in Score the model predictions. The actual rating is denoted by m, where m is the sample size.

[0039] Real ratings The calibration is performed in sections based on the actual condition of the mold and the dimensional deviation ΔD: Brand new condition: New mold ; Early wear: No defective products, and dimensional deviations. , ,in This represents the number of cycles already run. This refers to the number of cycles in the early wear stage, which accounts for 20% of the total lifespan, with a scoring range of 80-100. Mid-term wear and tear: Occasional non-conforming products. , ,in This refers to the number of cycles in the mid-term stage, accounting for 50% of the total lifespan, with a scoring range of 50-80. Late wear: Frequent non-conforming products , ,in The number of cycles in the late stage accounts for 30% of the total lifespan, with a scoring range of 0-50. Failure status: .

[0040] It should be further explained that transfer learning selects a trained mold model with the same material, process type, and similar structure as the source domain model, with a training data volume of ≥5000 cycles. It freezes 50%-70% of the feature mapping layer, the first 1-2 layers and the first 3 layers of the decision tree nodes, and trains only 30%-50% of the output layer of the target mold model. It inputs a small amount of labeled data of the target mold for more than 500 cycles, which can reduce the training data requirement of the target mold by 60%-70% and shorten the training time by 50%-60%.

[0041] It should be further explained that, based on the predicted health status score H, the mold condition is divided into four levels, including: health status Normal production requires no intervention; Attention status Reduce the offline size detection cycle by 50%, and pay attention to trend changes; Alert status Prepare spare parts and develop a preliminary maintenance plan; Failure risk status : Develop a detailed maintenance plan immediately.

[0042] It should be noted that the dynamic evolution prediction model is equipped with a model update triggering mechanism. When the prediction error continuously exceeds the threshold or the production materials are changed, the model parameters are automatically started to incrementally learn using the latest production data.

[0043] It should be further explained that the model update is triggered by the following condition: the prediction error exceeds a threshold for 3-5 consecutive periods, i.e. ,in =3 to 8 points; when production materials, process parameters, or mold repairs are updated, input 300-800 cycles of new data (X). new , ), update parameters according to incremental learning rules , where θ is the parametric model. For incremental learning rate, The gradient of the loss function for the new data is calculated; the parameters are updated, and the updated parameters are validated using an independent test set.

[0044] It should be specifically noted that the adaptive decision engine is implemented through a multi-objective optimization model. The optimization objective function of this model simultaneously considers production costs, product quality risks, equipment safety risks, as well as production context information such as the urgency of current production orders, spare parts inventory status, and available maintenance personnel information. By solving this function, a comprehensive cost dynamic maintenance strategy is obtained.

[0045] It should be further explained that the core of the adaptive decision engine is the objective function F, which is defined as: Where w1 to w6 are weight coefficients, and satisfy the following conditions: The general priority is recommended to be set as production cost weight. =0.3 to 0.5, quality risk weight =0.15 to 0.3, safety risk weight =0.15 to 0.3, order urgency weight =0.05 to 0.15, spare parts inventory weight =0.02 to 0.08, maintenance personnel availability weight =0.02 to 0.08, C is the production cost, R q For quality risks, R s For safety risks, E represents order urgency, S represents spare parts inventory, and M represents... p To ensure availability for maintenance personnel.

[0046] It should be further explained that the production cost ,in Maintenance costs are calculated based on industry averages, according to the type of maintenance. Downtime cost is the product of downtime duration and the equipment's output per unit time. The cost of scrapping is equal to the product of the defect rate, output, and unit cost of the product; quality risk. The lower the health status score, the higher the quality risk; the safety risk R s Segmented assignment, when R s =0.1, when R s =0.5, when R s =1.0; Order urgency ,in For order delivery time, For the current time, Total production cycle for the order; Spare parts inventory S, which is 1 when spare parts are needed for maintenance, and 0 otherwise; Maintenance personnel availability M. p The value is 1 when there are qualified and available maintenance personnel, and 0 otherwise.

[0047] It should be further explained that when the security risk R s When the value is 1.0, the safety risk weight w3 is forcibly increased to 0.4-0.5, and the order urgency weight w4 is decreased to 0.02-0.05; when the order urgency E≥0.8 and H≥50, w4 is increased to 0.15-0.2, and the maintenance cost weight w1 is decreased to 0.2-0.3.

[0048] It should be further explained that the adaptive decision engine uses a genetic algorithm to solve for the minimum value of the objective function F, with the following parameters set: population size 50-100; number of iterations 80-150; crossover probability 0.7-0.9; mutation probability 0.05-0.15; and the solution outputs the following three types of adaptive strategies: Immediate preventative replacement: Applicable or The aim is to completely eliminate the risk of failure and avoid production interruptions and safety accidents; Delay to the next scheduled maintenance window: Applicable to In addition, in scenarios with high order urgency, it is important to avoid disrupting the normal production cycle while ensuring a certain safety margin. Adjusting subsequent production process parameters to extend mold life (downgraded use): Applicable to Furthermore, in cases of insufficient spare parts inventory, reducing process parameters such as injection pressure and speed can slow down the mold wear rate, thus buying time for spare parts preparation.

[0049] S4: Strategy Output and Execution: Output dynamic maintenance strategies to user terminals for visualization, and convert dynamic maintenance strategies into control commands and send them to the production management system to automatically trigger corresponding maintenance execution actions.

[0050] It should be specifically noted that the conversion of dynamic maintenance strategies into control instructions involves automatically generating and issuing a corresponding standardized set of control instructions to the production management system based on the specific type of the dynamic maintenance strategy; immediately executing preventive replacements corresponds to generating preventive maintenance work order creation, dispatch, and resource reservation instructions; adjusting subsequent production process parameters corresponds to generating production equipment process parameter adjustment instructions.

[0051] It should be further noted that the strategy visualization display on the MES terminal provides the following visualization content: Mold Overview Status: Mold ID, production equipment number, and health status score are dynamically displayed on a color dashboard. Green indicates: ;yellow: ;orange color: ;red: The remaining lifespan is predicted based on the current H score and the total lifespan model, with the unit being the number of production cycles and the real-time production status indicator.

[0052] Strategy Details: Strategy Type, displays three types of strategies generated by the decision engine, priority, divided into high, medium, and low levels based on the objective function F value, and specific execution suggestions.

[0053] Risk and cost analysis: Show the quality risks associated with this strategy. Security risks And estimate production costs C, and provide quantitative comparisons of different strategies.

[0054] Historical analysis: H change trend and key parameter curves over the past 30 periods.

[0055] It should be further explained that the control command generation and issuance automatically generates standardized JSON commands based on the selected strategy, and then distributes them to various execution systems via the industrial bus network, as shown in the following example: Immediate preventive replacement instruction: The instruction content includes the instruction type "preventive maintenance execution", mold ID, equipment ID, required execution time, and detailed maintenance task list; associated system instructions include issuing to the injection molding machine to execute the shutdown instruction; issuing to the MES system to automatically generate a maintenance work order; and issuing to the spare parts management system to trigger the required spare parts issuance process.

[0056] Adjusting process downgrades using instructions: The instruction content includes the instruction type "process parameter adjustment", mold ID, equipment ID, instruction effective time, and a specific list of parameter adjustments; related system instructions include issuing to the injection molding machine controller to update the production formula; issuing to the quality inspection system to adjust the sampling frequency; issuing to the quality inspection system to adjust the sampling frequency.

[0057] It should be further explained that the execution system returns a receipt confirmation receipt within 10-30 seconds after receiving the instruction; after the instruction is executed, the system returns the execution result feedback; all feedback information is updated to the MES terminal interface in real time, forming a complete decision-making-execution-feedback closed loop.

[0058] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting mold life and formulating maintenance strategies based on big data analysis, characterized in that, include: S1: Data Acquisition and Fusion: Collect and correlate multi-source heterogeneous data in the mold production process. The multi-source heterogeneous data includes time-series data of process parameters from production equipment, time-series data of online monitoring from sensors, and structured and image data of offline inspection. S2: Collaborative Feature Engineering: Process the fused multi-source heterogeneous data to generate a structured training dataset with the production cycle as the basic unit; S3: Intelligent Prediction and Decision Making: The structured training dataset is input into a dynamic evolution prediction model to obtain the health status score of the mold's remaining useful life; the health status score, together with the current production context information, is input into the adaptive decision engine to calculate and generate a dynamic maintenance strategy. S4: Strategy Output and Execution: Output dynamic maintenance strategies to user terminals for visualization, and convert dynamic maintenance strategies into control commands and send them to the production management system to automatically trigger corresponding maintenance execution actions.

2. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 1, characterized in that: The multi-source heterogeneous data in the associated mold production process are associated and fused using mold ID and production cycle ID as unique identifiers. The process parameter timing data includes injection pressure, injection speed, and mold temperature timing data. The online monitoring time-series data includes online monitoring time-series data from vibration sensors and acoustic emission sensors; The structured and image data includes wear images of key parts of the mold and structured dimensional measurement data from the machine vision system.

3. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 1, characterized in that: The collaborative feature engineering process involves processing the fused multi-source heterogeneous data, including temporal alignment and collaborative feature extraction.

4. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 3, characterized in that: The time-domain alignment uses a single production cycle as the basic time unit, and maps online monitoring time-series data and process parameter time-series data with different sampling frequencies to each production cycle in a unified manner through timestamp alignment and data interpolation.

5. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 3, characterized in that: The collaborative feature extraction extracts time-domain statistical features and frequency-domain spectral features from online monitoring time-series data, and extracts visual features representing surface texture, crack length and area from image data. The extracted multivariate features are then fused with the process parameter features within the corresponding production cycle to construct a cycle-feature matrix as a structured training dataset.

6. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 1, characterized in that: The dynamic evolution prediction model is constructed using an ensemble learning algorithm and employs transfer learning technology to transfer the feature extraction layer of the trained model to the prediction model of the new model.

7. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 6, characterized in that: The dynamic evolution prediction model is equipped with a model update trigger mechanism. When the prediction error exceeds the threshold continuously or the production materials are changed, the model parameters are automatically started to incrementally learn using the latest production data.

8. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 1, characterized in that: The adaptive decision engine is implemented through a multi-objective optimization model. The optimization objective function of this model simultaneously considers production costs, product quality risks, equipment safety risks, as well as production context information such as the urgency of current production orders, spare parts inventory status, and available maintenance personnel information. By solving this function, a comprehensive cost dynamic maintenance strategy is obtained.

9. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 1, characterized in that: The dynamic maintenance strategy includes immediate preventative replacement, delaying execution to the next planned maintenance window, and downgrading use by adjusting subsequent production process parameters to extend mold life.

10. The method for predicting mold life and formulating maintenance strategies based on big data analysis according to claim 9, characterized in that: The process of converting dynamic maintenance strategies into control instructions involves automatically generating and issuing a corresponding set of standardized control instructions to the production management system based on the specific type of the dynamic maintenance strategy; immediately executing preventive replacements and generating instructions for creating, dispatching, and reserving preventive maintenance work orders; and adjusting subsequent production process parameters and generating instructions for adjusting production equipment process parameters.

Images