Industrial network intrusion detection using a multilayer perceptron: Experimental evaluation on the tennessee eastman process

Industry 4.0 has ushered in a new era of smart technologies, enabling interconnected systems powered by advancements such as artificial intelligence, big data, cloud computing, and industrial control systems (ICS). These technologies aim to streamline automation across every stage of ICS. However, these integrations introduce significant cybersecurity risks. This exposure makes production and process systems vulnerable to sophisticated cyberattacks that threaten operational integrity. This is why continuous monitoring of critical infrastructures is imperative to maintain seamless operations, particularly in the domain of supervisory control and data acquisition systems. Nevertheless, the integration of these systems with communication networks renders them vulnerable to potential intrusions by individuals harbouring malicious intentions. Existing statistical topologies often suffer from limited detection accuracy, slow response times, and inadequate handling of persistent cyberattack scenarios in ICSs. This research focuses on the detection of anomalies in ICSs exercising deep neural networks algorithm. An innovative neural network architecture based on multilayer perceptrons with six hidden layers is introduced. The proposed model reveals enhanced computational efficiency and reduced training time compared to other existing neural network architectures. It endorses a high accuracy score of 96.97 % and area under the curve percentage of 96.88 % in detecting cyberattacks and anomalies. The efficacy of the proposed anomaly detection framework (ADF) is assessed by subjecting it to two explicit types of cyberattacks; a ping of death and a fuzzification of an input flow value in the presence of disturbances. For dataset collection, training, and validation of ADF, a hybrid testbed is implemented, comprising the Tennessee Eastman process as an ICS benchmark integrated with an Advantech (Formula presented) PLC as a real-time hardware-based controller enabled with Modbus data communication protocol.