      APPLICATION OF NEURAL NETWORKS TO CONDITION BASED MAINTENANCE: A CASE STUDY IN THE TEXTILE INDUSTRY  Stefano Ierace, Roberto Pinto, Sergio Cavalieri   University of Bergamo, Department of Industrial Engineering Viale Marconi, 5 - I - 24044 Dalmine (BG), Italy E-mail: (stefano.ierace, roberto.pinto, sergio.cavalieri)@unibg.it   Abstract: The development of complex and sophisticated equipment makes necessary to enhance modern maintenance management systems. Conventional Condition-Based Maintenance (CBM) reduces the uncertainty of maintenance according to the needs indicated by the equipment condition. This paper describes a novel predictive maintenance system for textile machine systems based upon a neural network approach. This approach avoids the need for costly measurement of system parameters. The obtained results lead to the conclusion that neural networks represent an effective tool in supporting CBM policies. Copyright © 2002 IFAC  Keywords: Condition Based Maintenance (CBM), Predictive Maintenance, Artificial Intelligence, Neural Network, Fault-detection method    1. INTRODUCTION  Due to the even greater pressures on efficiency gains and quick response to customers’ requests, manufacturing industries are striving their efforts to reduce and eliminate costly, unscheduled downtimes and unexpected breakdowns. In this context, it is evident how a proper maintenance strategy can represent an important leverage for achieving such challenging results. As mentioned by Han and Yang (1996), maintenance costs can represent from 15% to 40% of the overall cost of goods sold, depending on the specific industry (Mobley, 1990). Another study conducted by Spiewak (2000) shows that one minute of downtime in an automotive manufacturing plant could amount up to $20.000. Moreover, the indirect impact of maintenance on operations and logistics activities (in terms of production costs, product quality, responsiveness, flexibility, etc.) can be substantial. As Lee et al. (2006) state, most companies still need to undertake a breakthrough in carrying out their maintenance activities by shifting from a traditional ”Fail and Fix (FAF)” maintenance approach to a ”Predict and Prevent (PAP)” maintenance methodology:  the first approach is purely reactive (“Fix it when it brakes”); no action is taken to prevent failures or to detect the imminent occurrence of a failure (Kothamasu et al, 2006);  the second one subsumes a clear separation of tasks between the production and the maintenance department in a “I operate – you fix” fashion (Waeyenbergh and Pintelon, 2002). The most acknowledged proactive approach is based on the Condition Based Maintenance (CBM) theory, which aims to signal and highlight a prompt detection and diagnosis of any deviations from the nominal operations of machines and the identification of the root cause stressor(s) responsible for this condition. This is a decision making strategy where the decision to perform maintenance is reached by observing the condition of the system and/or its components (Kothamasu et al, 2006). CBM is defined as ”maintenance carried out according to need as indicated by condition monitoring” (British Standard, 1984). In these areas, during the last decade, soft computing techniques as artificial neural networks (ANNs), genetic algorithms (GAs) or fuzzy logic have attracted a great deal of interest: as an example, Yam et al. (2001) used a neural network to carry out a reliable fault diagnosis in a power plant; Gao & Ovaska (2001) used genetic algorithms for motor fault diagnosis. This paper reports at first an overview of the most prominent neural network applications within the maintenance field. Starting from a case study in the textile industry, the paper aims also to provide a further contribution in the scientific literature to the acknowledgment of the capability of neural networks as predictive tools as a support to diagnostics and CBM strategies. The remainder of the paper is organized as follows: section 2 provides a brief description of the problem discussed, while section 3 reports a review on neural network techniques and its applications in maintenance. Section 4 describes the proposed