Vulnerability Mining Of Industrial Control Protocol Based On Deep Adversarial Learning

After the recent financial crisis,there occurred some new trends in the development of manufacturing.Governments have made strategic plans to build the next generation of manufacturing such as Industrial Internet and Chinese Manufacturing 2025.These strategies intend to empower the industry through information technology,such as optimizing the processes,reducing the costs and increasing the efficiency,Thus it can unlock greater productivity.Since the industrial internet is often related to safety-critical industry,ensuring the security of the industrial internet has special meaning.To facilitate collaboration between subsystems,more and more interconnections between different subsystems are established in the industrial internet.Thus,the system will face more external security threats.Before applying the system into actual production,it is necessary to discover potential system vulnerabilities in time and prevent them in advance.Currently,applying traditional fuzz testing techniques to discover loopholes in industrial control systems is an effective method.However,there are some limitations on applying the techniques.(i)High demand for the testers.The tester is required to design appropriate test cases according to the communication protocol specification running in the system.(ii)Long test cycle.The entire testing process will last a long time.it is impossible to complete the test task efficiently when it is in urgent needs.(iii)Not universal.Traditional methods design specific test cases based on specific test objectives.This paper proposes a fuzz testing case generation method,based on deep adversarial learning,to make up for the above limitations.First,it capture massive communication data frames from the target system.Second,it need train the model with the obtained data to get a specific model.Third,it generate a large amount of test case data with the generated model and stress test the system with the cases.Finally,improve the system according to the occurred abnormality.Experimental results in actual environment prove that the method has good performance.High test case pass rate can be obtained in different industrial control system.It can effectively cause abnormal behaviors of the system.The expected results are achieved in terms of test efficiency and test target independence.