Scholarlink

<p>Lubricating oil is crucial in mechanical systems, particularly in automobiles. Regular evaluation of engine oils is essential because deteriorated oils can cause engine damage and affect environmental sustainability. Research on oil contamination involves costly equipment and expert analysis. This study aimed to develop a low-cost multisensor system to detect lube oil impurity levels using machine-learning algorithms. The system uses a bare tapered bent multimode optical fiber (BTBMOF) sensor, MQ135 gas sensor, thermistor, infrared (IR) transmitter-receiver, and DHT11 sensor to detect oil parameters, such as temperature, viscosity, gas components, and particles. Data from these sensors were fed into the gradient boosting (GB), Bayesian inference (BI), random forest, support vector machine, logistic regression, and multilayer perceptron models to predict oil impurities. GB had the highest accuracy (83.56%). This study examined the interactions between the sensor outputs in detecting impurities. Excluding data for the gas, IR, and BTBMOF sensors relative to temperature reduced the accuracy to 63.01%, 83.56%, and 78.08%, respectively. The global SHapley Additive exPlanations (SHAP) analysis is applied to measure each individual feature's influence on model predictions. Optical signal shows the highest mean (|SHAP value|) for discrimination, followed by sample temperature. The IR and gas sensors show lower importance, while classification primarily uses optical and thermal data. Simultaneous data on oil temperature, viscosity, particulate matter, and trapped gases are needed to fully measure oil impurity levels. Machine learning algorithms serve as expert systems for predicting impurities. Thus, this machine-learning-based multisensor system can be used as a unified system for lubricating oil impurity measurements.</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

Novel instrumentation for in-situ determination of impurity level in engine oil using machine learning

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Rashmi Rekha Roy,
  • Sandip Bordoloi

摘要

Lubricating oil is crucial in mechanical systems, particularly in automobiles. Regular evaluation of engine oils is essential because deteriorated oils can cause engine damage and affect environmental sustainability. Research on oil contamination involves costly equipment and expert analysis. This study aimed to develop a low-cost multisensor system to detect lube oil impurity levels using machine-learning algorithms. The system uses a bare tapered bent multimode optical fiber (BTBMOF) sensor, MQ135 gas sensor, thermistor, infrared (IR) transmitter-receiver, and DHT11 sensor to detect oil parameters, such as temperature, viscosity, gas components, and particles. Data from these sensors were fed into the gradient boosting (GB), Bayesian inference (BI), random forest, support vector machine, logistic regression, and multilayer perceptron models to predict oil impurities. GB had the highest accuracy (83.56%). This study examined the interactions between the sensor outputs in detecting impurities. Excluding data for the gas, IR, and BTBMOF sensors relative to temperature reduced the accuracy to 63.01%, 83.56%, and 78.08%, respectively. The global SHapley Additive exPlanations (SHAP) analysis is applied to measure each individual feature's influence on model predictions. Optical signal shows the highest mean (|SHAP value|) for discrimination, followed by sample temperature. The IR and gas sensors show lower importance, while classification primarily uses optical and thermal data. Simultaneous data on oil temperature, viscosity, particulate matter, and trapped gases are needed to fully measure oil impurity levels. Machine learning algorithms serve as expert systems for predicting impurities. Thus, this machine-learning-based multisensor system can be used as a unified system for lubricating oil impurity measurements.