The introduction of fieldbus network architecture provided new chances and challenges to the modern manufacturing systems. The fieldbus networks improved the interchangeability of devices, provided the flexibility and reconfigurability of the ap plications and unproved the reliability of the systems, while reducing installation and maintenance costs. However, it also introduces different, forms of uncertainties between controllers, sensors arld actuators, as well the problems of maintenance in the networked environments. The main purpose of this doctoral study is to establish a systematic framework of performance assessment and fault prediction in networked industrial automation systems, through a general multi-layer strategy. In this study, the research focuses on the development of a general framework of intelligent maintenance system, concerning (1) the performance assessment and prediction of industrial network, (2) the fault diagnosis of networked automation systems, and (3) network bus-off hitting time estimation. In chapter III, a, network health monitoring tool is developed, in which the performance features extracted from network are used to quantitatively evaluate performance of the network. Physical layer features, such as power voltage, signal-to-noise ratio, etc., are selected. Diagnosis and prediction of network faults can be performed based on these features using a new similarity measure developed in this study. Scenarios of network degradations are investigated and results have shown the effectiveness of the diagnosis method. In chapter IV, in intermittent connection (IC) diagnostic method is developed that integrates the logic behavior (timing, sequence, etc.) with physical layer information obtained from the network. A new graph based IC isolation method is developed. The experiments conducted show that the developed method is able to locate the intermittent connection problems. Also in chapter IV, a quantitative method to estimate bus-off hitting time of the network node is developed. Using the IC diagnostic method developed earlier in the chapter, a methematic model for node transmitter error counter dynamaics is constructed based on observed error patterns. The experiments show the true bus-off hitting times agree well with the estimated values.