SF35 Deret Multi-Sensor IoT Berbasis AI untuk Termal Mesin Sepeda Motor

Abstract

Engine overheating is a primary cause of performance loss and premature component failure in motorcycles, yet factory temperature indicators are reactive and only warn once an unsafe condition has occurred. This study extends a prior IoT thermal-monitoring system by adding an artificial-intelligence (AI) layer and framing the platform as a teaching model for sustainable engineering education. A three-sensor array—PT100 (via MAX31865), Thermocouple Type-K (via MAX31855), and a non-contact GY-906 (MLX90614)—is read by an ESP32, displayed on a 20x4 LCD, and streamed to a Blynk dashboard with a buzzer, push, and email Early Warning System. Using 81 field samples collected from a Honda PCX 150 across daytime, evening, and night sessions with and without coolant, a lightweight edge-AI model (logistic regression) performs sensor fusion to (a) predict overheating two minutes ahead and (b) detect coolant loss before overheating occurs. Under leave-one-session-out cross-validation the predictor reached 93.8% accuracy and an AUC of 0.984, issuing warnings on average 2.0 minutes earlier than a fixed 100 C threshold, while cooling-fault detection identified coolant loss about 5 minutes before overheat onset. The IoT link stayed stable (latency < 2 s, zero packet loss). Beyond the technical result, the system is presented as a project-based learning module that integrates IoT, machine learning, and green technology to build sustainability-oriented graduate competencies.