This paper focuses on the remediation of stealthy adversaries perpetrating cyber-physical attacks. Machine learn- ing is to be applied to cover the verification of data and the discovery of the falsification of events for malicious purposes. Adversarial learning will be addressed as well. The work builds upon the concept of adversarial-learning adaptation and moving target defenses via control and artificial intelligence theory. The required challenge is to distinguish intentional attacks and component faults. Both need different reactions. The comparison of normal system operation with abnormal behaviour is the basis of all detection anomalies. We are currently exploring the possibility of recording the sound of the testbed, a fishertechnick (model of an industrial mini-factory ). To differentiate between abnormal and normal sound of each engine of the testbed.