| Numerous applications in complex systems are either theoretically intractable or hard to be solved within a practical time frame. Researchers are forced to implement a different intelligence like heuristics to solve problems. However, people, even experts, make mistakes under uncertain situations. That is, people often deviate from the rules of decision theory, which provides a set of compelling principles or desiderata for how people should behave. Thus, people use decision support models like expert systems to support their decisions and reduce mistakes that might lead to severe operational and economical losses. By encoding expert knowledge in a decision-theoretic framework, we can reduce errors in reasoning and thereby build decision support systems that offer recommendations of higher quality.;Our research focuses on developing an effective framework of the human and machine collaboration for fault diagnosis and detection in complex systems in order to generate satisfactory solutions. We refer to this as Human and Machine Co-Investigate Intelligence System (HM-CII). Although HM-CII framework could be used widely where the human and machine intelligences are involved in problem solving, we apply HM-CII in the area of fault diagnosis and detection which plays an important role in complex systems. With the current development of efficient data mining algorithms and the growing accessibility to a vast amount of data, the HM-CII model can be supported by automated access to data from existing data sources.;However, because of the different mechanisms of intelligence, human experts may not be able to read results based on these decision support systems easily. In addition, since the decision process in statistical and computational techniques depends on data that have been previously collected and updated, the information derived from the data analysis may sometimes be insufficient and noisy, and it can be difficult to track the evolution of the data structure. On the other hand, humans use their knowledge and experiences to shape and implement the decision-making process for fault diagnosis and prediction. Although these heuristics are quite useful, sometimes they lead to severe and systematic errors. This is especially true when the system becomes so complex that human experts reach the limits of their ability to analyze the system. We provide an alternative design of human and machine process by presenting an HM-CII methodology that addresses collaboration of human and machine intelligence. The proposed HM-CII does not just combine two processes for improved solutions but comes up with the agreement between humans and machines through several numbers of interactions. We investigate and report the confrontations and negotiations in the iterations. The HM-CII has two phases: (1) causal structure construction process and (2) interaction process of human and machine to select the best causal structure using rank aggregation process. In order to develop the model that fits various data structures, several alternative steps in HM-CII are presented with examples of ergonomic injury source detections and pipe-installation project delay detection problems. Finally, through the fault classification results, we show the potential for improvement in the quality of solutions that can be achieved using the HM-CII approach. |
