CN-122022538-A - Intelligent alloy quality diagnosis system

CN122022538ACN 122022538 ACN122022538 ACN 122022538ACN-122022538-A

Abstract

The intelligent alloy quality diagnosis system provided by the invention realizes multi-source data fusion and intelligent decision of the whole alloy smelting process by constructing a collaborative optimization framework integrating data acquisition, processing, analysis and service application, can organically integrate quality diagnosis, fault early warning and process optimization, and forms closed-loop control by feeding back an optimization strategy to a production process control system in real time, thereby remarkably improving the quality stability of alloy products, the self-adaptation capability of the process and the overall optimization level of production benefits.

Inventors

ZHANG YUAN
ZHAO HONGSHENG
WANG XIN
GAO XIANG
WANG HUAYANG
ZHENG HENGJU
ZHAO ZIYI
XIAO HE
ZHU ZHILIN
LIU HONGBING

Assignees

鄂尔多斯市蒙泰铝业有限责任公司

Dates

Publication Date: 20260512
Application Date: 20251205

Claims (10)

1. An intelligent alloy quality diagnostic system, comprising: the data acquisition module is used for acquiring multi-source heterogeneous data in real time through a sensor network deployed in the alloy smelting equipment, and carrying out anomaly identification and data enhancement processing on the multi-source heterogeneous data to generate an enhanced data stream; The data processing module is connected to the data acquisition module and is used for receiving the enhanced data stream, cleaning the enhanced data stream, extracting features and carrying out standardization processing on the enhanced data stream and outputting a standardized feature set; The intelligent analysis module is connected to the data processing module and is used for receiving the standardized feature set, processing the standardized feature set through the quality diagnosis sub-module, the fault prediction sub-module and the process optimization sub-module and generating an analysis result comprising alloy components, performance diagnosis results, smelting process fault prediction alarms and smelting process optimization strategies; The application service module is connected to the intelligent analysis module and used for receiving the analysis result and providing visual display, early warning notification and decision support of the analysis result; The intelligent analysis module dynamically updates model parameters according to smelting data fed back by the production process control system to form closed-loop optimization, wherein the production process control system controls the alloy smelting equipment and is used for executing process parameter setting and collecting smelting data.
2. The system of claim 1, wherein the data acquisition module comprises high precision temperature sensors, voltage sensors, current density probes, spectrometers for component analysis, and furnace condition anomaly collectors for deployment at key process points of a smelting furnace.
3. The system of claim 1, wherein the data processing module comprises a feature extraction sub-module for performing multi-scale feature engineering, in particular comprising extracting frequency domain features using a short-time fourier transform, obtaining time-frequency domain features using a wavelet transform, and extracting depth features from sensor image data.
4. The system of claim 1, wherein the intelligent analysis module comprises a multi-modal heterogeneous data fusion sub-module that uses a framework of a neural network and an attention mechanism to construct a dynamic graph structure to aggregate information and output fusion feature vectors.
5. The system of claim 4, wherein the intelligent analysis module further comprises an adaptive migration learning sub-module that migrates source domain model parameters to target domains through task similarity metrics based on a model independent meta-learning framework.
6. The system of claim 5, wherein the process optimization sub-module generates the smelting process optimization strategy by a pre-set reinforcement learning algorithm that relies on a multipart optimization objective function for training and decision, the method of constructing the optimization objective function comprising the steps of: Acquiring quality diagnosis features and energy consumption features, and calculating to obtain a first optimization target component based on a plurality of alloy quality index evaluation values output by the quality diagnosis submodule and combining the deviation condition of real-time energy consumption data acquired from the data acquisition module relative to historical reference energy consumption; Acquiring process state characteristics, and calculating to obtain a second optimization target component based on the proximity degree between the current values of the standardized plurality of key smelting process state parameters output by the data processing module and the corresponding process specification target values; Generating and quantifying adjustment action characteristics, generating a plurality of smelting process parameter adjustment quantity suggestions by the process optimization submodule, carrying out constraint evaluation on the amplitude of each adjustment quantity according to historical statistical data of each adjustment quantity, and calculating to obtain a third optimization target component; And integrating and calculating a comprehensive optimization target value, and integrating the first optimization target component, the second optimization target component and the third optimization target component to generate an overall evaluation value for guiding the reinforcement learning algorithm to find the optimal technological parameters.
7. The system of claim 6, wherein in calculating the first optimization objective component, a dynamic weight coefficient is assigned to each alloy quality indicator evaluation value, and the method for determining the dynamic weight coefficient includes the steps of: Based on historical batch data analysis, obtaining stability performance data of each alloy quality index in historical production and difference data of relevant production cost and an expected target thereof, and comprehensively calculating to obtain a weight basic value; Based on the analysis result output by the multi-mode heterogeneous data fusion sub-module, acquiring the influence intensity data of a plurality of key process influence parameters on the specific alloy quality index under the current production working condition, and combining the historical reference influence intensity, and calculating to obtain a weight adjustment value; and integrating the weight basic value and the weight adjustment value, and finally determining a dynamic weight coefficient corresponding to each alloy quality index for calculating the first optimization target component through normalization processing.
8. The system of claim 1, wherein the intelligent analysis module further comprises a privacy-preserving federal learning sub-module, the privacy-preserving federal learning sub-module is configured to train the model locally at the plurality of clients by combining homomorphic encryption with differential privacy, and to perform encryption aggregation of model parameters at the server.
9. The system of claim 1, wherein the intelligent analysis module further comprises an interpretable artificial intelligence sub-module that performs causal reasoning based on the structural equation model and generates a counterfactual interpretation to provide a decision basis.
10. The system of claim 1, wherein the application service module comprises an early warning notification sub-module, configured to send out an alarm by mail, a short message, and a mobile application push mode when the probability confidence of the smelting process failure output by the failure prediction sub-module exceeds a preset threshold.

Description

Intelligent alloy quality diagnosis system Technical Field The invention relates to the technical field of industrial intelligent manufacturing, in particular to an intelligent alloy quality diagnosis system. Background Along with the continuous development of industrial intelligent manufacturing technology, quality control and process optimization of an alloy smelting process become key links for improving product quality and reducing production cost. Traditional alloy smelting process monitoring mainly relies on detection of a single data source and an alarm mechanism based on a fixed threshold value, key process parameters are acquired through deployment of basic sensors, and quality assessment and process adjustment are carried out by combining an empirical model. However, the traditional mode has obvious limitations that each functional module is relatively independent, the data flow and the analysis process are mutually split, end-to-end closed loop optimization from data acquisition to process execution cannot be realized, so that the quality diagnosis, the fault prediction and the process adjustment lack of cooperative linkage, and the complex requirements of multi-parameter coupling and multi-objective optimization in the smelting process are difficult to adapt, thereby influencing the quality stability and the production benefit of the final product. Disclosure of Invention In view of the above, the invention provides an intelligent alloy quality diagnosis system to solve the technical defects existing in the prior art. Specifically, the invention provides an intelligent alloy quality diagnosis system, which comprises: The data acquisition module is used for acquiring multi-source heterogeneous data in real time through a sensor network deployed in the alloy smelting equipment, and carrying out anomaly identification and data enhancement processing on the multi-source heterogeneous data to generate an enhanced data stream; the data processing module is connected to the data acquisition module and is used for receiving the enhanced data stream, cleaning the enhanced data stream, extracting the characteristics and carrying out standardized processing on the enhanced data stream and outputting a standardized characteristic set; The intelligent analysis module is connected to the data processing module and used for receiving the standardized feature set, processing the standardized feature set through the quality diagnosis sub-module, the fault prediction sub-module and the process optimization sub-module and generating an analysis result comprising alloy components and performance diagnosis results, a smelting process fault prediction alarm and a smelting process optimization strategy; the application service module is connected to the intelligent analysis module and used for receiving the analysis result and providing visual display, early warning notification and decision support of the analysis result; The intelligent analysis module dynamically updates model parameters according to smelting data fed back by the production process control system to form closed-loop optimization, wherein the production process control system controls alloy smelting equipment and is used for executing process parameter setting and collecting smelting data. In some embodiments, the data acquisition module includes high precision temperature sensors, voltage sensors, current density probes, spectrometers for component analysis, and furnace condition anomaly collectors for deployment at key process points of the smelting furnace. In some embodiments, the data processing module includes a feature extraction sub-module for performing multi-scale feature engineering, including in particular extracting frequency domain features using a short-time fourier transform, acquiring time-frequency domain features using a wavelet transform, and extracting depth features from sensor image data. In some embodiments, the intelligent analysis module includes a multi-modal heterogeneous data fusion sub-module, which constructs a dynamic graph structure to aggregate information and output fusion feature vectors using a framework of combining a graph neural network and an attention mechanism. In some embodiments, the intelligent analysis module further comprises an adaptive migration learning sub-module that migrates the source domain model parameters to the target domain through the task similarity metric based on the model independent meta-learning framework. In some embodiments, the process optimization sub-module generates a smelting process optimization strategy through a preset reinforcement learning algorithm, the algorithm is trained and decided by relying on an optimization objective function formed by multiple parts, and the construction method of the optimization objective function comprises the following steps: Acquiring quality diagnosis features and energy consumption features, and calculating to obtain a first optimization