基于LOF的联合收获机制造质量检测与分级系统研究

LOF-based Combine Harvester Manufacturing Quality Detection and Grading System

摘要:

随着制造业对于产品质量的要求越来越高，机器学习技术在制造质量控制中的应用开始受到关注。针对联合收获机制造质量检测过程自动化和集成化程度较低、缺乏定量评价手段等问题，设计开发了一套联合收获机制造质量终检系统，在此基础上提出了“终检系统+二次分级”的制造质量混合检测方法，通过终检软件排查合格区间以外的异常数据，筛选劣质产品；通过分级模型对合格产品进行二次检测，标记质量隐患。在整合和分析联合收获机制造质量检测需求的基础上提出了检测流程并通过Visual Components数字车间仿真平台对总体方案进行仿真和测试。根据实际需求和检测功能开发了基于LabVIEW平台的联合收获机终检系统软件，并设计了人机交互界面。试验结果表明系统可以满足各项检测需求并实现产品质量检测功能，初步验证了系统可行性。结合使用场景选用局部异常因子（Local outlier factor，LOF）作为二次分级算法，根据异常检测原理将其集成到检测流程中，并建立了制造质量检测与分级算法架构，依据处理结果将初筛合格的产品二次分类并标记为“good”和“tracked”，进而完善制造过程质量检测-评价体系。训练结果表明LOF可以在差异性不显著的数据集中识别异常样本，性能验证过程中该方法可以准确识别并标记测试数据集中的“tracked”样本，且与四分位图的分布一致，进一步验证了该混合检测方法的有效性。本研究开发的联合收获机制造质量检测系统和提出的分级方法具有应用价值，将数字车间架构与机器学习方法应用于农机装备产品制造质量检测，为复杂农机装备制造质量控制提供了解决思路和方法。

Abstract:

With the increasing demand for product quality in the manufacturing industry, the application of machine learning （ML） technology in manufacturing quality control has been under attention. To address the low automation and integration, as well as the lack of quantitative evaluation methods in the manufacturing quality inspection for combine harvester, a combine harvester manufacturing quality end-of-line inspection system was designed and developed. Based on this system, an "end-of-line inspection + secondary grading" manufacturing quality hybrid inspection method was proposed, which used the inspection software to screen out abnormal products outside the qualified range and select superior and inferior products. The secondary grading model performed a secondary inspection on qualified products and marks hidden problems. Firstly, based on the integration and analysis of the combine harvester manufacturing quality inspection requirements, the detection flow was designed. The overall design of the system was tested and simulated by using the Visual Components digital workshop platform. The LabVIEW-based end-of-line inspection software was developed according to the actual requirements and detection functions, and corresponding userfriendly human-machine interfaces were designed. The results of the end-of-line workshop inspection tests showed that the system can meet various inspection requirements and achieve software functions, preliminarily verifying the feasibility of the system. Secondly, local outlier factor （LOF） was selected as the secondary grading algorithm according to the scenario, and it was integrated into the detection flow based on its anomaly detection principle. Then, a manufacturing quality inspection and grading framework was established, and the grading process classified the initially screened qualified products into "good" and "tracked" groups based on the processing results, thereby improving the manufacturing quality inspection and evaluation system. The training results indicated that LOF-based method can identify anomalous samples in the dataset with insignificant differences. In the performance validation process, this method accurately identified the four "tracked" samples in the testing dataset, which was consistent with the distribution of the quartile plots, further validating the effectiveness of this hybrid detection method. The developed end-of-line inspection system for the manufacturing quality of combine harvesters and the proposed grading method had important practical application value, promoting the application of digital workshop concept and ML on agricultural machinery, and providing solutions and methods for agricultural machinery manufacturing quality control.