Застосування машинного навчання в інтелектуальній системі керування ковшем навантажувача

Abstract. Problem. Wheel loaders are versatile machines widely used for loading, unloading and transporting various materials. Modern trends in the improvement of these machines focus on increasing the level of their automation. Automation reduces the operator’s workload, enhances safety, and improves the productivity and energy efficiency of machines under various operating conditions. One of the tasks of wheel loader automation is controlling bucket loading. Due to the complex interaction of the bucket with soil or other loaded material and the influence of numerous uncertain dynamic factors, the use of model-free machine learning methods in bucket control systems is reasonable. Goal. The goal of this paper is to analyze machine learning methods in terms of their suitability for developing an intelligent bucket control system for wheel loaders. Methodology. We examined three main types of machine learning – supervised, unsupervised, and reinforcement learning with a detailed evaluation of reinforcement learning methods such as Q-learning, Deep Q-Network, REINFORCE, and Proximal Policy Optimization. Then, we compared these methods based on flexibility, implementation complexity, adaptability, and data dependency to assess their suitability for an intelligent bucket control system. Results. The analysis identifies the Actor-Critic approach as the most suitable, with its potential for combination with other methods highlighted to enhance system performance in dynamic operating conditions. Originality. Through a systematic comparison of several reinforcement learning algorithms, we identified the Actor-Critic approach as the most effective method for controlling the loader bucket. Practical value. The results of the study will help to improve the efficiency of intelligent loader bucket control system.