Aluminum profile machining quality detection system and method

Abstract

The application belongs to the technical field of intelligent manufacturing, and discloses an aluminum profile machining quality detection system and method, which comprises the following steps: acquiring time sequence process data and corresponding finished product performance data in the production process; training a data processing model through a multi-modal contrast learning method, mapping data of different modes to a unified implicit representation space, and realizing deep semantic association between the process and the final performance in the space; and mapping process data of the aluminum profile to be detected to the space, and evaluating the product quality through the relative position of the representation vector to the preset standard. The application solves the problems of heterogeneous data fusion difficulty, attribution process black box and control lag, and improves the intelligent level of quality control.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing technology, specifically relating to an aluminum profile processing quality inspection system and method. Background Technology

[0002] The consistency of aluminum profile quality is crucial to the success of their application in key fields such as aerospace, automotive, and construction. Their final performance is determined by the entire complex production process (extrusion, quenching, aging). Traditional quality control methods, such as endpoint sampling or surface defect detection based on a single data source, suffer from significant lag and cannot reveal the complex nonlinear relationship between process fluctuations and internal performance evolution.

[0003] With the development of artificial intelligence technology, some methods have attempted to correlate process parameters and product quality using machine learning models. However, in the process of realizing this invention, the inventors discovered at least the following problems in the prior art:

[0004] 1. The production process involves various heterogeneous data, such as one-dimensional time-series sensor signals (temperature, pressure), two-dimensional images of the finished product's internal structure (metallographic images), and mechanical property data in the form of function curves (stress-strain curves). Existing methods typically simply stitch these together or process them independently, failing to establish deep semantic relationships between cross-modal data, i.e., failing to truly understand how a temperature curve specifically affects the texture of a metallographic image.

[0005] 2. When a model predicts a quality defect, it typically only provides the correlation weights of the input parameters, failing to offer an intuitive, physically-intuitive explanation. Engineers cannot see how an abnormal process parameter gradually leads to the deterioration of the microstructure, thus limiting their ability to quickly locate and resolve the problem.

[0006] 3. Existing quality control methods are mostly reactive or offline optimization based on static models. For sudden, minor disturbances during the production process, the system lacks the ability to assess the future impact of these disturbances on the quality of the final product in real time, thus making it impossible to conduct accurate and timely in-process intervention or proactive control. Summary of the Invention

[0007] The present invention aims to at least partially solve the aforementioned technical problems. Therefore, the objective of the present invention is to provide a system and method for inspecting the processing quality of aluminum profiles.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for inspecting the processing quality of aluminum profiles includes: acquiring time-series process data and / or finished product performance data corresponding to the time-series process data during the aluminum profile production process; training a data processing model based on a multimodal contrastive learning method to map the time-series process data and the finished product performance data to a unified implicit representation space, such that the distance between the process data representation and the performance data representation from the same aluminum profile sample in the implicit representation space is less than the representation distance from different aluminum profile samples; inputting real-time or historical time-series process data of the aluminum profile to be inspected into the data processing model to obtain its process trajectory or endpoint representation in the implicit representation space, and generating a product quality inspection result based on the relative position of the trajectory or endpoint representation with a preset standard area.

[0010] Preferably, the time-series process data includes time-series data of at least one of temperature, pressure, and speed; the finished product performance data includes at least one of metallographic micrographs, mechanical property curves, and hardness distribution maps.

[0011] Preferably, the method further includes training a conditional generator decoder to generate visual attribution data for one or more modalities, taking vectors in the implicit representation space as input. Examples of visual attribution data include predicted metallographic micrographs of the finished product, predicted mechanical property curves, or simulated process parameter curves that lead to anomalies. When the quality inspection result is abnormal, the implicit space representation corresponding to the abnormal state is input into the conditional generator decoder to generate visual attribution data related to the anomaly.

[0012] Preferably, the method further includes continuously mapping real-time process data to an implicit representation space to form a real-time trajectory during the production process, and performing forward prediction and feedforward control when the trajectory deviates from the preset golden trajectory area.

[0013] Accordingly, the present invention also provides an aluminum profile processing quality inspection system, comprising: a data acquisition unit; a unified representation fusion unit, which integrates a data processing model trained through multimodal contrastive learning, the model including a process encoder for processing time-series process data and a performance encoder for processing finished product performance data, for mapping different modal data to a unified implicit representation space; and a quality assessment unit for comparing the implicit representation of the sample to be inspected with a preset standard.

[0014] Preferably, the system further includes: a generative attribution unit, which integrates a conditional generation decoder for providing visualized multimodal attribution reports; and a dynamic control unit for implementing forward-looking process control based on real-time trajectory analysis.

[0015] Finally, the present invention provides a computer-readable storage medium storing an executable program that, when executed, implements any of the foregoing methods.

[0016] The beneficial effects of this invention are as follows:

[0017] This invention constructs a unified implicit representation space through multimodal contrastive learning, and for the first time connects process data and final performance characterization data at the semantic level, enabling the model to understand rather than fit the relationship between them, thereby fundamentally improving prediction accuracy and generalization ability.

[0018] With the help of a generative decoder, the system can translate any abstract quality problem or process deviation into intuitive multimodal data (such as predicted metallographic images), realizing the transformation from a "black box" to a "white box" and providing engineers with a powerful, interactive causal analysis tool.

[0019] By performing trajectory analysis in the implicit space, the system can predict the impact of minor fluctuations in the current process on the future in real time, and make precise and minimal feedforward interventions, thus elevating quality control from a passive response to a proactive prevention approach. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the offline training phase in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the system architecture and data flow during the online application phase in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] It should be understood that, and also noted, in the embodiments, the functions / actions may appear in a different order than those shown in the figures. For example, depending on the functions / actions involved, they may actually be performed substantially concurrently, or sometimes the two figures shown consecutively may be performed in reverse order.

[0024] like Figure 1 and Figure 2As shown, this invention provides an aluminum profile processing quality inspection system, which is implemented through two stages: offline training and online application. The system mainly includes a data acquisition unit for acquiring data, a unified representation fusion unit for deep data fusion, and a quality assessment unit for judging product quality. The following section will describe the collaborative methods of each unit in conjunction with the steps of aluminum profile processing quality inspection.

[0025] like Figure 1 The offline training phase, as shown, aims to train the core data processing model within the unified representation fusion unit to construct a unified representation space that profoundly reflects the intrinsic relationship between "process and performance." First, the data acquisition unit obtains a large number of paired samples from historical production data. Each pair of samples contains two parts: one part is time-series process data, such as sensor readings of temperature, pressure, and speed during extrusion and quenching processes of aluminum profiles; the other part is the strictly corresponding finished product performance data, such as metallographic micrographs of finished product slices or mechanical property curves obtained from tensile tests. This data is fed into the unified representation fusion unit, which integrates a process encoder and a performance encoder. During training, the system employs a multimodal contrastive learning method. Specifically, the time-series process data and finished product performance data are mapped into high-dimensional process representation vectors and performance representation vectors, respectively, through their respective encoders. The model's goal is to minimize the distance between process representation vectors and performance representation vectors from the same production sample within a unified implicit representation space, while maximizing the distance between vector pairs from different production samples. In this way, the model learns to align and fuse data from two different modalities at the semantic level.

[0026] The online application phase is the application process of the system in actual production, and its system architecture and data flow are as follows: Figure 2 As shown. When quality inspection of a new aluminum profile is required, the data acquisition unit first acquires the real-time or historical process data of the profile. Then, this data is fed into the pre-trained process encoder in the unified representation fusion unit to obtain its process trajectory or endpoint representation in the implicit representation space. This representation is then passed to the quality assessment unit, whose core function is... Figure 2 The implicit spatial trajectory analysis module in the system implements this. This module compares the relative position of the obtained representation vector with a preset standard area consisting of a large number of qualified product representation vectors, and generates product quality inspection results accordingly. If the representation vector falls within the standard area, the system will determine it as qualified in the quality report output; otherwise, it will determine it as unqualified or at risk.

[0027] Furthermore, preferred embodiments of the present invention also include more advanced functional modules to provide deeper analysis and more proactive control, such as... Figure 2As shown, the system may include a generative attribution unit. This unit is activated when the implicit spatial trajectory analysis module detects an anomaly and can generate multimodal visual attribution reports (such as predicted deterioration metallographic images). These reports can ultimately be presented to the engineer's interactive interface to help the engineer intuitively understand the root cause of the problem.

[0028] Additionally, the system may include a dynamic control unit. This unit analyzes the process trajectory in real time, makes forward-looking predictions when deviations are detected, and generates feedforward control commands. These commands then proactively intervene in upstream processes through the production line control system, thereby achieving preventative quality control.

[0029] This invention is not limited to the above-described optional embodiments. Anyone can derive other various forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that falls within the scope of the claims of this invention shall be protected by this invention.

Claims

1. A method for inspecting the processing quality of aluminum profiles, characterized in that, Includes the following steps: Obtain the time-series process data and / or the finished product performance data corresponding to the time-series process data during the aluminum profile production process; Based on the multimodal contrastive learning method, a data processing model is trained to map the time-series process data and the finished product performance data to a unified implicit representation space, such that in the implicit representation space, the distance between the process data representation and the performance data representation from the same aluminum profile sample is smaller than the representation distance from different aluminum profile samples. The real-time or historical time-series process data of the aluminum profile to be tested is input into the data processing model to obtain its process trajectory or endpoint representation in the implicit representation space, and the product quality inspection result is generated based on the relative position of the trajectory or endpoint representation and the preset standard area.

2. The method according to claim 1, characterized in that, The time-series process data includes time-series data of at least one of temperature, pressure, and speed; the finished product performance data includes at least one of metallographic micrographs, mechanical property curves, and hardness distribution maps.

3. The method according to claim 1 or 2, characterized in that, The method also includes: Train a conditional generator decoder that can generate data in at least one modality corresponding to any vector in the implicit representation space as input; When the quality inspection result is abnormal, the implicit space representation corresponding to the abnormal state is input into the conditional generation decoder to generate visual attribution data related to the abnormality in one or more modalities.

4. The method according to claim 1, characterized in that, The method also includes: During the production process, the real-time acquired time-series process data is continuously mapped to the implicit representation space to form a real-time process trajectory. When the real-time process trajectory deviates from a preset golden trajectory area, the future direction of the trajectory and the product performance corresponding to its endpoint are predicted. Based on the difference between the predicted performance and the target performance, a feedforward control instruction is generated and executed to adjust the production process parameters so that the real-time process trajectory returns to the golden trajectory region.

5. The method according to claim 3, characterized in that, The visualized attribution data includes predicted metallographic micrographs of the finished product, predicted mechanical property curves, or simulated process parameter curves that lead to anomalies.

6. A quality inspection system for aluminum profile processing, characterized in that, include: The data acquisition unit is used to acquire time-series process data and finished product performance data during the aluminum profile production process; The unified representation fusion unit integrates a data processing model trained through multimodal contrastive learning, which is used to map the data from different modalities to a unified implicit representation space. The quality assessment unit is used to compare the representation of the aluminum profile to be tested in the implicit representation space with a preset standard and output the quality test result.

7. The system according to claim 6, characterized in that, The system also includes: The generative attribution unit integrates a conditional generation decoder to receive anomaly state representations from the quality assessment unit and generate multimodal visual attribution reports.

8. The system according to claim 6 or 7, characterized in that, The system also includes: The dynamic control unit is used to monitor the process trajectory generated by the unified representation fusion unit in real time, make forward predictions when the trajectory deviates, and generate feedforward control commands to be sent to the production equipment.

9. The system according to claim 6, characterized in that, The unified representation fusion unit includes a process encoder for processing time-series process data and a performance encoder for processing finished product performance data.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images