At present, the security situation in the industrial internet is becoming more and more serious. Various threats such as network attacks, malicious code and vulnerability utilization are gradually increasing. Consequently, it is urgent to study industrial threat detection methods. In order to tackle typical network attacks, system vulnerabilities and malicious operations, a real-time intelligent industrial threat detection method is proposed by analyzing the network data in the industrial control system. Particularly, artificial intelligence technique, adversarial sample generation technique and deep learning model are used in the method. Besides, the proposed method is achieved in the real network, and the corresponding industrial threat detection platform is developed. The results show that the developed threat detection platform can detect a variety of typical network attacks, system vulnerabilities, malicious code, etc. At the same time, the platform has good throughput and compatibility and is suitable for the actual industrial environment.