Scholarlink

<p>In recent years, advancements in sensor technologies, wired and wireless communications, computing, and artificial intelligence have significantly enhanced our ability to monitor, diagnose, and predict the health of vehicles. The increasing complexity, automation, and intelligence of vehicle systems have amplified the demand for predicting vehicle conditions. Recognizing this trend, numerous automotive companies have shown a keen interest in leveraging vehicle Diagnostic Trouble Code (DTC)-related sensor data for advanced diagnosis and prediction, aiming to enhance customer satisfaction and service efficiency. This study addresses the challenge of predicting DTC occurrences using collected DTC-related sensor data from commercial vehicles. To tackle this issue, we employ three Deep Learning (DL) methods for anomaly detection: LSTM-Autoencoder, 1D CNN-LSTM-Autoencoder, and Anomaly Transformer. To enhance the efficiency of DTC precursor time prediction, we propose a heuristic method incorporating sliding window detection and outlier detection within a multivariate ReConstruction Error (RCE) control procedure based on the best DL method. The performance of our proposed method is discussed using case examples. Finally, we deliberate on the limitations of our study and suggest potential directions for future research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning-Based Vehicle DTC Prognosis Approach

  • 期刊论文

  • 发表日期：

    2025年6月11日

查看全文
  • Hong-Bae Jun,
  • Hansom Kim,
  • Saeyan Eom,
  • Seonghyun Jeon,
  • Seungbum Ha,
  • Sewoong Jung,
  • Beomkyu Park

摘要

In recent years, advancements in sensor technologies, wired and wireless communications, computing, and artificial intelligence have significantly enhanced our ability to monitor, diagnose, and predict the health of vehicles. The increasing complexity, automation, and intelligence of vehicle systems have amplified the demand for predicting vehicle conditions. Recognizing this trend, numerous automotive companies have shown a keen interest in leveraging vehicle Diagnostic Trouble Code (DTC)-related sensor data for advanced diagnosis and prediction, aiming to enhance customer satisfaction and service efficiency. This study addresses the challenge of predicting DTC occurrences using collected DTC-related sensor data from commercial vehicles. To tackle this issue, we employ three Deep Learning (DL) methods for anomaly detection: LSTM-Autoencoder, 1D CNN-LSTM-Autoencoder, and Anomaly Transformer. To enhance the efficiency of DTC precursor time prediction, we propose a heuristic method incorporating sliding window detection and outlier detection within a multivariate ReConstruction Error (RCE) control procedure based on the best DL method. The performance of our proposed method is discussed using case examples. Finally, we deliberate on the limitations of our study and suggest potential directions for future research.