Industrial Plants such as refineries are complex installations that require constant monitoring and inspection to ensure safe and stable operation. This is currently conducted by field operators and one of the important tasks is acoustic inspection, i.e., listening for abnormal sounds while on patrol, skills of which are difficult to describe in operation procedures. Due to the issues of skilled staff retirement, costs, and inspection quality variance, the automation of acoustic inspection is desirable. Due to the large scale of these installations, several acoustic landscapes co-exist. This makes the establishment of a single model for abnormal sound detection difficult. Therefore, considering a mobile robot patrolling the plant, this study proposes to divide the robot's path into a grid where in each grid cell a distinct model is trained, bypassing the issue of differing acoustic landscapes. Experiments conducted in a simulated environment confirmed the effectiveness of the proposed method.