AUTOMATION OF PROCESSES IN TRANSPORT ELECTROMECHANICAL SYSTEMS BASED ON IOT TECHNOLOGIES

Abstract

The purpose of the study is to develop an integrated solution for monitoring the condition and adaptive control of traction electric drives based on IoT and edge computing technologies in order to improve the efficiency, reliability, and controllability of vehicle electric drive systems. Methodology. At the deployment stage, a sensor infrastructure with deep sleep mode and short wake-up periods is created. Inside the vehicle, critical high-speed signals are transmitted via a CAN bus with minimal delay, even when high-level ECUs are unavailable. Each unit is equipped with a wireless node based on a microcontroller with Wi-Fi, which measures the temperature in the bearing and winding areas using precise digital sensors, records vibrations with MEMS accelerometers, and measures phase or bus currents with the Hall effect. Static and dynamic calibration is used to ensure data reliability. Telemetry is preprocessed on board with filtering and standardization by z-scores, and spectral analysis is performed for vibrations to detect signs of bearing defects and mechanical resonances. Key indicators are generated locally, and in case of suspicious events, short fragments of raw signals are transmitted to the cloud via MQTT with buffering. The speed loop is implemented by a PI controller with anti-windup and parameter projection, and an actor-critic reinforcement learning agent, refined in simulation, updates the coefficients within safety limits. A compact neural network reduces overshoot and smoothes torque, while edge MPC coordinates regenerative braking with local constraints. The cloud level aggregates streams from multiple machines, performs anomaly detection, estimates residual resource, and plans motion profiles and recovery shares at the fleet level. Scientific novelty. A multi-level architecture of an integrated monitoring and control system is proposed, combining local data processing at the node level with peripheral intelligence, adaptive control of electric drive parameters using reinforcement learning, and coordination of recuperation processes at the cloud level. The developed solution ensures effective interaction between sensor, computing, and control components without a significant increase in computing resources. Conclusions. As a result of experimental and field tests, more than 97% of packets were delivered at a distance of more than 30 m, with a temperature measurement accuracy of about ±0.1 °C, detection of imbalances, bearing defects, and transient current overloads. Software and hardware modeling demonstrated a reduction in transient processes, integral error, and increased resistance to parametric shifts. In real conditions, up to 18% of kinetic energy was recovered, and service costs were reduced by 20–25%, confirming the effectiveness and practical applicability of the proposed system.