The immune system in a living organism serves the purpose of defending the organism from external intrusions, such as microbes, viruses or foreign bodies. The immune system identifies the presence of an intrusion, isolates the intrusion and attempts to eliminate it, or at least eliminate its effects on the normal body function. Improper function or absence of the immune system causes serious problems and often death (e.g., leukemia, cancer or AIDS). In this talk, several approaches to realizing immune system like functions of anomaly detection, fault isolation and self-healing in complex engineered systems will be discussed.

Firstly, multistage manufacturing systems with autonomous compensation of dimensional errors, using an explicit process model and distributed measurement and actuation points, will be discussed. Practical implications of sensor and actuation point placement will be addressed from the point of newly introduced theoretical concepts of diagnosability and compensability, analogous to the concepts of observability and controllability in control theory. Both deterministic and stochastic analyses of the problem will be presented.

Secondly, a novel approach to anomaly detection and fault isolation in complex dynamic systems, such as automotive or aircraft engines, will be presented. In the case of such highly sophisticated systems, the traditional approach to diagnostics becomes excessively cumbersome because of the need to train the condition monitoring units to recognize a large number of faults under highly diverse control and environmental conditions, some (most) of which often cannot be even anticipated in advance. The newly proposed approach is inspired by natural immune systems and achieves fault detection and isolation through distributed anomaly detection. This is realized through a new modeling approach for identification of non-linear dynamic systems, where the operational space of the system is divided through unsupervised clustering into regions, within which system behavior can be modeled using simpler, analytically more tractable, models. Theoretical findings will be presented regarding the dependency of model convergence and accuracy on the topological structure of the model. Implications of local model tractability to realizing self-healing functionality through adaptive control will also be discussed.

Dragan Djurdjanovic obtained his B.S. in Mechanical Engineering and in Applied Mathematics in 1997 from the University of Nis, Serbia; his M.Eng. in Mechanical Engineering from the Nanyang Technological University, Singapore in 1999; his M.S. in Electrical Eng. (Systems) and Ph.D. in Mechanical Eng. in 2002 from the University of Michigan. His research interests include advanced quality control in multistage manufacturing systems, intelligent proactive maintenance techniques and applications of advanced signal processing in biomedical engineering. He co-authored 25 journal publications, one book chapter and 18 conference publications. He is the recipient of several prizes and awards, including the 2006 Outstanding Young Manufacturing Engineer Award from the Society of Manufacturing Engineers (SME), 2005 Teaching Incentive Award from the Dept. of Mechanical Eng. of the University of Michigan, Nomination for the Distinguished Ph.D. Thesis from the Dept. of Mechanical Eng., University of Michigan in 2003, and The Outstanding Paper Award at 2001 SME North American Manufacturing Research Conference.