Symbolic Representation of Sensor Data and Its Application in Fault Diagnosis for TE Process

This paper presents an efficient fault diagnosis framework for the Tennessee Eastman process by integrating symbolic data representation with information-theoretic analysis. To handle high-dimensional temporal data, this study employs Symbolic Dynamics and Symbolic Aggregate Approximation to convert continuous sensor data into discrete symbolic sequences. Symbolic Transfer Entropy is then utilized to extract discriminative features that capture dynamic information flow from these sequences. A Linear Discriminant Analysis classifier is finally applied for fault identification. Experimental results demonstrate that the method achieves an excellent balance between accuracy and speed: the combination of Symbolic Dynamics and Symbolic Transfer Entropy attains a high diagnosis accuracy, while the combination of Symbolic Aggregate Approximation and Symbolic Transfer Entropy reduces computational time significantly. This work confirms the strong potential of the approach for practical industrial applications requiring both reliability and efficiency.