Optimized deep learning approach for automated fault diagnosis in mobile robot used for fire-fighting application

Robots are mechanical devices with human-like appearances that carry out difficult jobs. Mobile platforms can expand the workspace compared to fixed-based robots, whose capacity is constrained, and it is used in various applications. Firefighting robots are essential for switching off fire without human presence, and it is mostly placed in ruthless environments where they are consistently vulnerable to many faults. If a fault is not diagnosed in time, it can cause unexpected downtime and even catastrophic damage to the machinery. However, some existing algorithms, such as You Only Look Once, Version 3 (Yolov3), STMicroelectronics32 (STM32), and Haar Cascade Classifier, are used for fault detection, but their accuracy is limited when detecting multiple faults. In order to solve this problem, a deep learning-based LSTM model was used to improve fault detection accuracy in mobile robots. Initially, a fire detection system for a mobile robot platform was developed using smoke, flame, and temperature sensors. The system used relays, 1n5822 diodes, an IRF3205 power Metal Oxide Semiconductor Field Effect Transistor (MOSFET), capacitance, and ULN 2003 type relay drivers for defect detection. During simulations, the fault detection model collects data from the sensors. Then, the acquired data is sent for pre-processing using Min-Max normalization to normalize the data within the particular range. Then, the pre-processed data is fed into Optimized Long Short-Term Memory (O-LSTM) to diagnose fault in robotic sensors. Bald eagle search (BES) optimization is utilized in the created sources to determine the learning rate, number of hidden layers, and node in order to reduce error for enhancing deep learning prediction accuracy. Performance indicators for proposed and existing models are compared in order to assess the planned model's performance. Performance metrics such as accuracy, recall, specificity, and precision attained for the proposed model is 97%, 90%, 95%, and 97%. Through this proposed O-LSTM model occurrence of fault in firefighting mobile robots can be detected more effectively.