密封电子设备多余物属性识别模型的可靠性评估与可解释性分析

doi:10.19595/j.cnki.1000-6753.tces.251327

Abstract
Figure/Table
References
Related Citation (15)

Download: PDF (6808 KB) HTML (1 KB)
Export: BibTeX | EndNote (RIS)

Abstract The loose particles generated during the production process of sealed electronic equipment pose a severe threat to their operational reliability. The particle impact noise detection method can detect the presence of loose particles; it cannot identify their attributes, such as location and material. Recognizing the attributes of loose particles has become a hot topic. In practical applications, a single detection task typically involves multiple detection events, and existing methods lack the mechanism to convert the classification results from these events into a single recognition result for the task. The impact of loose-particle attribute recognition models at the detection-event level on recognition accuracy at the detection-task level remains unclear. In addition, the contribution of each functional module in loose particle attribute recognition models needs to be evaluated to explain the synergistic effect between performance reliability and recognition stability.
This paper proposes a framework for reliability evaluation and interpretability analysis of loose particle attribute recognition models. Firstly, by introducing majority voting, the classification results for all detection events in a single detection task are aggregated, and the category with the highest frequency is determined as the attribute recognition result. Thus, a conversion mechanism from event classification to task recognition is established, and the issue of converting attribute recognition results in practical applications is solved. Secondly, to evaluate the robustness of the loose particle attribute recognition models, a novel definition of recognition reliability has been proposed. That is, under the same operating conditions, when the model's performance shows reasonable degradation, as long as its global classification performance and the local classification performance for each category remain above 0.5, the accuracy of attribute recognition results can still be guaranteed through majority voting. Thus, the impact of regularity on the relationship between performance reliability and recognition accuracy is clarified. Then, cumulative progressive, individual progressive, and sequential failure analysis methods are used to evaluate the five functional modules during model construction. It is shown that pulse processing and feature construction are the two key functional modules with the greatest impact on model performance.
Finally, numerous experiments in real-world sealed electronic equipment detection scenarios have shown that the loose particle attribute recognition models can consistently achieve global and local classification performance above 0.5 under varying operating conditions. Therefore, accurate and reliable location and material recognition results can be continuously produced through majority voting, demonstrating the proposed framework's practicality and robustness. This paper provides a reference for the reliability evaluation and robustness application of machine learning models.

Key words： Loose particle attribute recognition model reliability assessment interpretability analysis majority voting detection reliability key functional modules

Received: 24 July 2025

PACS:

TM930

	Service

	E-mail this article
	Add to my bookshelf
	Add to citation manager
	E-mail Alert
	RSS
	Articles by authors
	Sun Zhigang
	Wang Guotao
	Chen Hao
	Xiao Kaiwen
	Zhai Guofu

Cite this article:

Sun Zhigang,Wang Guotao,Chen Hao等. Reliability Evaluation and Interpretability Analysis of Loose Particle Attribute Recognition Model for Sealed Electronic Equipment[J]. Transactions of China Electrotechnical Society, 2026, 41(16): 5663-5680.

URL:

https://dgjsxb.ces-transaction.com/EN/10.19595/j.cnki.1000-6753.tces.251327 OR https://dgjsxb.ces-transaction.com/EN/Y2026/V41/I16/5663

[1] 陈必玉, 卢静, 余协正, 等. 航空航天产品多余物预防与控制研究[J]. 航空标准化与质量, 2024(6): 47-50.
Chen Biyu, Lu Jing, Yu Xiezheng, et al.Research on prevention and control of redundant materials in aerospace products[J]. Aeronautic Standardization & Quality, 2024(6): 47-50.
[2] 王衎, 耿海, 陆帆, 等. 宇航电推进产品多余物防控的认识与实践[J]. 质量与可靠性, 2024(1): 9-12, 20.
Wang Kan, Geng Hai, Lu Fan, et al.Cognition and practice of the redundancy prevention and control in the aerospace electric propulsion products[J]. Quality and Reliability, 2024(1): 9-12, 20.
[3] 任建文, 成钢, 李永春, 等. 航天电子产品多余物预防与控制方法探析[J]. 航天工业管理, 2023(12): 20-24.
Ren Jianwen, Cheng Gang, Li Yongchun, et al.Analysis on remainder prevention and control methods for aerospace electronic products[J]. Aerospace Industry Management, 2023(12): 20-24.
[4] 王淑娟, 王国涛, 翟国富, 等. 航天继电器多余物微粒碰撞噪声检测的冲击试验条件[J]. 电工技术学报, 2011, 26(1): 75-80, 85.
Wang Shujuan, Wang Guotao, Zhai Guofu, et al.Impact test condition of particle impact noise detection for space relay remainders[J]. Transactions of China Electrotechnical Society, 2011, 26(1): 75-80, 85.
[5] 李鹏飞, 翟国富, 孙志刚, 等. 密封电子元器件与装置多余物检测发展综述[J]. 电工技术学报, 2024, 39(10): 3152-3179, 3217.
Li Pengfei, Zhai Guofu, Sun Zhigang, et al.A review of the loose remainders detection for sealed electronic components and devices[J]. Transactions of China Electrotechnical Society, 2024, 39(10): 3152-3179, 3217.
[6] Zhang Guozheng, Li Menghui, Gu Xin, et al.Fault diagnosis method for open-circuit faults in NPC three-level inverter based on WKCNN[J]. CES Transactions on Electrical Machines and Systems, 2025, 9(2): 234-245.
[7] 丁志千, 翁凯迪, 杨福, 等. 基于泄漏信号特征参数分析法的储罐底板声发射检测[J]. 中国特种设备安全, 2025, 41(6): 13-19.
Ding Zhiqian, Weng Kaidi, Yang Fu, et al.Acoustic emission detection of tank bottom plate based on leakage signal characteristic parameter analysis method[J]. China Special Equipment Safety, 2025, 41(6): 13-19.
[8] Lei Yaguo, Yang Bin, Jiang Xinwei, et al.Appli-cations of machine learning to machine fault diagnosis: a review and roadmap[J]. Mechanical Systems and Signal Processing, 2020, 138: 106587.
[9] Feng Ke, Ji J C, Ni Qing, et al.A review of vibration-based gear wear monitoring and prediction techniques[J]. Mechanical Systems and Signal Processing, 2023, 182: 109605.
[10] Sun Zhigang, Liang Qi, Zhai Guofu, et al.Multi-source information fused loose particle localization and material identification method for sealed electronic equipment[J]. Engineering Applications of Artificial Intelligence, 2025, 139: 109529.
[11] Sun Zhigang, Wang Guotao, Zhai Guofu, et al.Method of locating loose particles inside aerospace equipment based on parameter-optimized XGBoost[J]. Engineering Applications of Artificial Intelligence, 2023, 126: 106796.
[12] 孙志刚, 王国涛, 高萌萌, 等. 基于kNN优化算法的密封电子设备多余物定位技术[J]. 电子测量与仪器学报, 2021, 35(3): 94-104.
Sun Zhigang, Wang Guotao, Gao Mengmeng, et al.Sealed electronic equipment loose particle positioning technology based on kNN algorithm of parameter optimization[J]. Journal of Electronic Measurement and Instrumentation, 2021, 35(3): 94-104.
[13] 孙志刚, 王国涛, 高萌萌, 等. 参数优化支持向量机的密封电子设备多余物定位方法研究[J]. 电子测量与仪器学报, 2021, 35(8): 162-174.
Sun Zhigang, Wang Guotao, Gao Mengmeng, et al.Research on localization method of loose particles inside sealed electronic equipment based on parameter-optimized support vector machine[J]. Journal of Electronic Measurement and Instrumentation, 2021, 35(8): 162-174.
[14] Sun Zhigang, Jiang Aiping, Gao Mengmeng, et al.Feature optimization method for the localization technology on loose particles inside sealed electronic equipment[J]. Expert Systems with Applications, 2022, 204: 117569.
[15] Sun Zhigang, Gao Mengmeng, Jiang Aiping, et al.Incomplete data processing method based on the measurement of missing rate and abnormal degree: Take the loose particle localization data set as an example[J]. Expert Systems with Applications, 2023, 216: 119411.
[16] Ye Xuerong, Chen Hao, Chen Cen, et al.Life-cycle dynamic robust design optimization for batch production of permanent magnet actuator[J]. IEEE Transactions on Industrial Electronics, 2021, 68(10): 9885-9896.
[17] Huang Hongzhong, Li He, Shi Yan, et al.Theory and application of possibility and evidence in reliability analysis and design optimization[J]. Journal of Reliability Science and Engineering, 2025, 1(1): 015007.
[18] 施亦非, 黄宇峰, 王锋, 等. 基于参数优化特征模态分解的强背景噪声下滚动轴承故障诊断[J]. 振动与冲击, 2024, 43(21): 107-115.
Shi Yifei, Huang Yufeng, Wang Feng, et al.Fault diagnosis of rolling bearing under strong background noise based on POFMD[J]. Journal of Vibration and Shock, 2024, 43(21): 107-115.
[19] 王攀攀, 李兴宇, 张成, 等. 基于角域重采样和特征强化的电机滚动轴承故障迁移诊断方法[J]. 电工技术学报, 2025, 40(12): 3905-3916.
Wang Panpan, Li Xingyu, Zhang Cheng, et al.Fault transfer diagnosis method for motor rolling bearings based on angular domain resampling and feature enhancement[J]. Transactions of China Electrotech-nical Society, 2025, 40(12): 3905-3916.
[20] Zhai Guofu, Sun Zhigang, Wang Guotao, et al.Four-channel pulse extraction and matching method for loose particle localization on sealed electronic equipment[J]. IEEE Transactions on Instrumentation and Measurement, 2025, 74: 3515017.
[21] 钱鑫, 王国涛, 王淑娟. 密封电子元件及设备多余物的PIND方法研究进展[J]. 航天制造技术, 2016(6): 1-5, 16.
Qian Xin, Wang Guotao, Wang Shujuan.Development of PIND detection methods for sealed electronic components and equipments[J]. Aerospace Manufacturing Technology, 2016(6): 1-5, 16.
[22] Cao Yuan, Sun Yongkui, Xie Guo, et al.A sound-based fault diagnosis method for railway point machines based on two-stage feature selection strategy and ensemble classifier[J]. IEEE Trans-actions on Intelligent Transportation Systems, 2022, 23(8): 12074-12083.
[23] 徐玉珍, 邹中华, 刘宇龙, 等. 基于多尺度卷积神经网络和双注意力机制的V2G充电桩开关管开路故障信息融合诊断[J]. 中国电机工程学报, 2025, 45(8): 2992-3003.
Xu Yuzhen, Zou Zhonghua, Liu Yulong, et al.Information fusion diagnosis of switching tube open-circuit fault in V2G charging piles based on multi-scale convolutional neural network and dual-attention mechanism[J]. Proceedings of the CSEE, 2025, 45(8): 2992-3003.
[24] 马速良, 武建文, 袁洋, 等. 多振动信息下的高压断路器机械故障随机森林融合诊断方法[J]. 电工技术学报, 2020, 35(增刊2): 421-431.
Ma Suliang, Wu Jianwen, Yuan Yang, et al.Mechanical fault fusion diagnosis of high voltage circuit breaker using multi-vibration information based on random forest[J]. Transactions of China Electrotechnical Society, 2020, 35(S2): 421-431.
[25] 黄杰, 李建闽, 朱冰凡, 等. 基于条件网络和知识注入的多标签学习电能质量扰动识别与分类方法[J]. 电工技术学报, 2025, 40(23): 7652-7663.
Huang Jie, Li Jianmin, Zhu Bingfan, et al.A multi-label learning method based on conditional networks and knowledge injection for power quality disturbance identification and classification[J]. Transa-ctions of China Electrotechnical Society, 2025, 40(23): 7652-7663.
[26] 胡俊杰, 陈泉希, 潘羿. 面向快速响应与可调能力恢复的储能系统实时调度-再调度策略[J]. 电工技术学报, 2026, 41(9): 3022-3037.
Hu Junjie, Chen Quanxi, Pan Yi.Real-time scheduling-rescheduling strategy of energy storage system for fast response and adjustable capacity restoration[J]. Transactions of China Electrotechnical Society, 2026, 41(9): 3022-3037.