Fresh Approaches for Structured Text Programmable Logic Controllers Programs Verification

Abstract

Programmable logic controllers (PLCs) are everywhere today and perform critical tasks in industries. They are considered as a key component for the Industry 4.0. Before they are put into operation, it is necessary to check the accuracy of the PLC programs. This verification operation can be performed using model checkers. This stage is often long and costly and requires a domain expert who can understand the system, as well as the different model checker tools able to verify the code implemented in the controller. Furthermore, this verification often requires a conversion of the PLC code into a language understood by a model checker which can influence the behavior of the observed PLC. Hence, there is a need to propose methods and tools which could be used by technicians and engineers. The aim of this paper is to propose methods that require little work to set up and are robust to program sizes used in Industry 4.0. This paper explores some fresh ideas for human-adapted PLC code verification. We present different methods to test codes in structured text (ST) compliant with the IEC 61131-3 standard. Hence, the first idea is to test the ST code that will be directly implemented on a controller. For that, we propose a method using the model checker UPPAAL which allows us to obtain exact results on short codes. Second, we propose verifying the generic properties that a PLC program must avoid: deadlocks, non-accessible states and fugitive states or actions. To solve combinatory explosion problems encountered with the UPPAAL software, the third proposition consists of using relational databases. The same verification as previously followed can be obtained, but the search time is longer. The fourth and last proposal is to process the ST code with a neural network composed of long short-term memory layers (LSTM) to quickly determine the validity of the code. This method could give an approximation of code errors in a few seconds. The different proposed methods are supported with several examples.