Abstract
The paper describes a development of a diagnosing and operating system in which mutual cooperation between a system and an operator can be effectively used in decision making. Firstly, a concept of cooperation between the system and the operator was proposed. The authors discuss the cooperative method for plant operation and diagnosis. Then, the authors designed a framework of this system which is composed of three modules, namely a diagnosing module, an interface module and a supporting tool module. In this system, a knowledge base is represented by utilizing a fault tree and describing additional information for cooperative decision making. The authors implemented a prototype of the system for diagnosing human errors and malfunctions in a bio-process system. Finally, the availability of the prototype system is shown through simulated system operation.
Introduction
Modern industrial plants are increasingly complex and highly automated. While automation improves efficiency, it often creates a "black box" effect where human operators struggle to understand the system's internal state. This leads to a double-edged sword: operators may make incorrect judgments that the system blindly follows, and traditional expert systems are often too rigid to handle qualitative data or situations outside their pre-programmed knowledge base.
This research moves away from traditional "user-led" or "system-led" architectures toward a cooperative model. By leveraging both human flexibility and machine calculative precision, the proposed system aims to manage sensitive bioprocesses more effectively than either could alone.
System Architecture
The researchers developed a prototype system consisting of three core modules designed to facilitate this human-machine synergy:
- Diagnosing Module: Utilizes an inference engine and a knowledge base to identify process anomalies.
- Supporting Tool Module: Features a fermentation simulator based on differential equations, allowing operators to perform "what-if" analyses in a virtual environment.
- Interface Module: A graphical user interface (GUI) that facilitates the exchange of information, displaying sensor data alongside diagnostic dialogue.
Knowledge Representation and Cooperative Diagnosis
The most innovative aspect of the system is its use of a modified Fault Tree (FT) structure for knowledge representation. Unlike standard fault trees, each node in this system is explicitly tagged as either "automatically judgeable by the system" or "requires human cooperation."
When the diagnostic engine reaches a "cooperative" node—such as determining if a manual valve was left open or if a misoperation occurred—it initiates a dialogue with the operator. This approach turns the human into a "mobile sensor," providing qualitative data that the automated system cannot detect on its own.
Experimental Evaluation
The system was implemented using G2 (a real-time expert system tool) and evaluated through simulated scenarios involving the cultivation of Rhodococcus bacteria. The authors built a Virtual Fermentor to generate high-fidelity synthetic data, including bacterial growth rates, limiting substrate mass, and product yield.
Key Results
- Diagnosis of Human Error: In a test case where an operator input an excessive vitamin dose (1.4 g/L instead of the optimal 1.04 g/L), the system successfully traced the drop in productivity back to a "human input error" by cross-referencing sensor data with operator feedback.
- Diagnosis of System Failure: The system identified mechanical issues, such as a failure in the yeast extract supply valve, by ruling out human error through dialogue and analyzing the physical state of the virtual plant.
- Decision Support: Using the supporting simulator, operators were able to estimate optimal dosages that maximized production, demonstrating the system's utility as a decision-support tool.
Discussion and Conclusion
The cooperative framework demonstrates that machines can overcome the limitations of sparse sensor data in bioprocesses by effectively communicating with human operators. The inclusion of a simulator acts as a safety net, encouraging operators to gain intuition through virtual trial and error.
However, limitations remain. The current model assumes human input is always accurate and does not yet account for "secondary human errors" during diagnostic dialogues. Future work will focus on handling simultaneous cascading failures and improving cognitive ergonomics to ensure the interface remains manageable during process crises.
Summary updated