Abstract: A detection device is provided in the disclosure. The device uses unsupervised or self-supervised neural networks to learn nominal conditions of a target system, such as a device or a machine. The trained neural networks can reproduce sensory signals of the target system as a neural-network-reconstructed version of the sensory signals in the nominal conditions of a target system. The equipment anomaly detection device may analyze and predict operation conditions of the target system based on the neural-network-reconstructed version exceeding a certain level. When signal difference between the sensory signals and the neural-network-reconstructed version exceeds a certain level, the equipment anomaly detection device may issue an alert signal to reflect abnormal operation conditions of the target system.

Abstract: A detection device is provided in the disclosure. The device uses unsupervised or self-supervised neural networks to learn nominal conditions of a target system, such as a device or a machine. The trained neural networks can reproduce sensory signals of the target system as a neural-network-reconstructed version of the sensory signals in the nominal conditions of a target system. The equipment anomaly detection device may analyze and predict operation conditions of the target system based on the neural-network-reconstructed version exceeding a certain level. When signal difference between the sensory signals and the neural-network-reconstructed version exceeds a certain level, the equipment anomaly detection device may issue an alert signal to reflect abnormal operation conditions of the target system.