Special issue on point of contact between bio-technology and precision engineering. Malfunction diagnosis of bioprocess.

Hajime Asama, Isao EndoJournal of the Japan Society for Precision Engineering 1990 · DOI

Published in 1990, this research explores the intersection of biotechnology and precision engineering, focusing on the then-emerging field of AI-driven malfunction diagnosis in bioprocesses. At the time, managing large-scale fermentation was a significant challenge due to the inherent complexity of living systems.

The Challenge of "Ill-Structured Systems"

Bioprocesses are notoriously difficult to automate because they are "ill-structured systems." Unlike chemical or mechanical processes, bioprocesses involve living organisms whose internal metabolic states often remain a "black box." The authors identified several hurdles: a lack of real-time sensors for critical physiological data, low sensor reliability in harsh bioreactor environments, and the non-linear complexity of biological parameters. Consequently, plant operators often relied on "expert intuition" rather than formal models to adjust parameters during malfunctions.

The BIOACS Framework

To address these challenges, the authors proposed a tiered approach starting with the Bio Advanced Control System (BIOACS). This foundational system was designed to improve data quality through online turbidity meters for monitoring cell growth and automated sampling systems for measuring substrate and product concentrations.

The Bio Advanced Control System (BIOACS) architecture, showing the integration of sensors and controllers.

Beyond hardware, the system utilized Kalman Filtering to estimate "specific rates" (such as growth and consumption rates) in real-time. This provided a clearer picture of the biological state than raw sensor data alone.

AI-Driven Diagnosis

The core of the proposal was an expert system architecture that combined different reasoning methods. The system used Frame-based Modeling to represent the physical structure of the bioreactor and Production Rules (if-then logic) to diagnose hardware failures and operator errors. These components were integrated into a Blackboard Model, allowing various "knowledge sources" to contribute to a shared diagnostic workspace. The system was implemented in Smalltalk/V, leveraging object-oriented programming for a flexible, network-like representation of knowledge.

Physiological Activity as a Benchmark

A key insight of the paper is that physical parameters like temperature and pH are insufficient for comprehensive diagnosis. Instead, one must monitor the rates of biological change. The authors utilized normalized specific rate curves for growth ($\mu^*$), substrate consumption ($\nu^*$), and product formation ($\pi^*$) as benchmarks.

Normalized specific rate curves for growth, consumption, and production used to detect biological anomalies.

By comparing real-time data against these historical norms stored in the Bio DataBase System (BDBS), the system could identify biological malfunctions—such as mutation or contamination—that would be invisible to standard PID controllers.

Results and Legacy

The authors successfully implemented a prototype on a Fujitsu M780 mainframe, incorporating 154 diagnostic rules and 213 frames. The system demonstrated the ability to detect deviations in lactic acid bacteria cultures that traditional systems would miss. While limited by the computational constraints of 1990, this work represents a pioneering application of "Good Old-Fashioned AI" (GOFAI) to the complex "wetware" of biology, moving toward a future of networked, intelligent control systems.

Summary updated

BibTeX
@article{asama1990special,
  author = {Hajime Asama and Isao Endo},
  title = {{Special issue on point of contact between bio-technology and precision engineering. Malfunction diagnosis of bioprocess.}},
  journal = {Journal of the Japan Society for Precision Engineering},
  year = {1990},
  doi = {10.2493/jjspe.56.239},
}
Figure