Humans demonstrate ingenious ways of solving engineering problems, but their approaches to finding faults are often laborious and time-intensive. Their manual techniques do not scale well, especially when the systems are large, complicated, or widely-distributed geographically. Multi-Agent Systems (MAS) are one way to address these concerns, creating software entities called \textit{agents} that can automate fault-finding tasks. Agents operate co-operatively and semi-autonomously, working in ways that are similar to how teams of humans might tackle a problem.

This research centres on fault identification and diagnosis for complex automation and control systems known as Industrial Cyber-Physical Systems (ICPSs). The IEC 61499 Function Block Standard is a reference software architecture that is well-suited to creating an ICPS. However, while current function block design tools provide limited design-time fault-finding capabilities, they cannot be used to facilitate thorough diagnostic investigations when problems occur later in the life cycle of an ICPS. Larger, safety-critical systems require a level of fault diagnosis that demands more comprehensive approaches; design and development-time testing techniques are not sufficient.

The thesis proposes that it is both feasible and worthwhile to implement fault diagnostics as part of a Model-Driven Development methodology for ICPS that are built from IEC 61499 Function Blocks. Furthermore, it demonstrates how the design and implementation of diagnostic functionality can become a core part of the engineering processes. It demonstrates how to manage the inevitable faults that ICPSs suffer. Diagnostics can and should be a key component of a mature design and development methodology. The research demonstrates how fault identification and management capabilities can be created in-parallel with the components and subsystems that will ultimately realise the stakeholders' vision for their application.

A Scoping Survey Literature Review was performed to uncover mature fault diagnostic approaches used in the aerospace, automotive and manufacturing domains. These sectors employ ICPSs in safety-critical environments. Hence it is expected that it is in these environments that the most well-refined and mature fault management approaches will have evolved. Following this survey, a Design Science Research plan was created to facilitate multiple, incremental experimental phases. The primary phase produced a Software Architecture Document for a Fault Diagnosis Engine (hereafter referred to as \lquote the engine\rquote) by applying an Attribute-Driven Design (ADD) methodology. This process identified which quality attributes would best address the stakeholder's requirements.

This led to the creation of novel software-in-loop telemetry capture devices called \textit{Diagnostic Points}. These operate inside the FORTE IEC 61499-compatible function block application runtime, passing telemetry to and from the GORITE Multi-Agent System framework. By implementing agents who work co-operatively to capture and interpret evidence of system misbehaviour, diagnostic techniques were evaluated to determine which approaches most accurately identify and diagnose faults.

Each subsequent Design Science phase explored and implemented part of the engine proposed in the Software Architecture Document. The scope and evaluation criteria for each Design Science phase are defined by and reference back to the relevant sections of this document. The on-going Architectural Trade-off Analysis Method (ATAM) evaluations of each phase were driven by a set of pre-defined scenarios which were based on typical real-world faults. These were uncovered in both the scoping survey and during the subsequent research. Three journal articles and two conference papers were published during the research to refine the work, informed by peer-reviewed feedback.

The completed engine was evaluated during the design and development of two example function block applications. Following a V-Model methodology, the agents performed diagnostic testing at appropriate development and implementation stages. The results from these projects led to further refinement of both the engine and the Model-Driven Development with Diagnostics methodology that the thesis proposes.

This research has shown that the engine has reached a level of maturity that is close to the NASA Technology Readiness Level 5 (TRL 5). This level designates technologies that have been demonstrated to work in a relevant or representative physical environment beyond the laboratory. The aspects of the technology that contribute towards meeting the TRL 5 criteria are explored in greater depth in the conclusions chapter.