Thermally induced displacements in machine tools cause defective parts and limit five-axis machining accuracy. Compensation methods can reduce these displacements with minimal hardware effort, but due to their complexity these methods are rarely used. ARX models show promise for adaptive compensation of displacements within the kinematic chain of a machine tool. Experiments show significant reduction in displacements across the five-axis workspace. Implementing ARX models on NC controls in production faces constraints like low acceptance of production interruptions. ARX models are complex black-box models, making verification difficult. This paper presents a modified model structure that simplifies implementation, improves prediction quality, and reduces sensor noise. It also addresses the problem of interruption of production by necessary measurements. The model is implemented on a TNC 640 control with a five-axis machine tool, with training and testing conducted in a climate chamber.