Scholarlink

<p>Predictive maintenance plays a pivotal role in Industry 4.0, especially on complex machinery, optimizing cost in due process. A proactive approach of predictive maintenance employs data analytics and artificial intelligence to spot possible issues before they cause an asset to fail, lessening the likelihood of expensive repairs and unforeseen downtime. This paper performs predictive maintenance on water pumps by estimating the remaining useful life using machine and deep learning approaches. A multitasking artificial neural network has been developed to simultaneously identify fault mode and the remaining useful life. Due to high amounts of null values in the Dataset, the multitask model was trained in a semi-supervised setting. The trained multitask model produced 99.73% accuracy for the task of machine status identification and Mean Absolute Error (MAE) of 18.15 for the Remaining Useful Life (RUL) prediction task. The proposed multitask algorithm can be highly effective for predictive maintenance of water pumps in real time.</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

Proactive maintenance for water pumps: multitask neural network for fault detection and RUL estimation

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Mohammed Akheel,
  • Aditya Anjanikar,
  • Soham Navale,
  • V. A. Sairam,
  • Preksha Pareek,
  • Shivali Amit Wagle,
  • Mohd. Aquib Ansari

摘要

Predictive maintenance plays a pivotal role in Industry 4.0, especially on complex machinery, optimizing cost in due process. A proactive approach of predictive maintenance employs data analytics and artificial intelligence to spot possible issues before they cause an asset to fail, lessening the likelihood of expensive repairs and unforeseen downtime. This paper performs predictive maintenance on water pumps by estimating the remaining useful life using machine and deep learning approaches. A multitasking artificial neural network has been developed to simultaneously identify fault mode and the remaining useful life. Due to high amounts of null values in the Dataset, the multitask model was trained in a semi-supervised setting. The trained multitask model produced 99.73% accuracy for the task of machine status identification and Mean Absolute Error (MAE) of 18.15 for the Remaining Useful Life (RUL) prediction task. The proposed multitask algorithm can be highly effective for predictive maintenance of water pumps in real time.