Machine learning is increasingly revolutionizing machining, particularly in the fields of quality control and process optimization. Autoencoder techniques show great promise for detecting voids and material anomalies, which are critical for enhancing manufacturing efficiency by reducing or identifying scrap parts. Moreover, machine learning lays the groundwork for automated process control in wood machining, as natural wood variability often complicates anomaly detection. This study introduces an analogy experiment using defined holes in quasi-homogeneous MDF to address these challenges. Acceleration and force data collected during milling were used to train an autoencoder to identify undisturbed paths, thus establishing a baseline for anomaly detection. Testing the model with datasets containing drilled holes demonstrated its capability to detect anomalies. This study highlights the autoencoder's effectiveness not only in identifying artificial defects but also in detecting natural wood anomalies, such as fibre misalignments. Yet, it also demonstrated that the selection of signals is of paramount importance. Nevertheless, the findings underscore machine learning’s potential for improving quality control in wood machining. By enabling accurate and efficient anomaly detection, this research work can pave the way for adaptive, intelligent process management, including multisensory solutions and the integration of measurement systems at the spindle for more complex machining paths.