A case study is presented, where a predictive maintenance solution for non-critical machinery (such as elevators and machine tools) was sought. Both cases are different. There is no experience in elevator monitoring and diagnosis, and modeling has been performed using Neural Networks. On the other hand, machine tools were monitored through vibration systems where some experience exists. In this case, Bayesian Networks are the paradigm of choice as it was also recommended to include some ‘adaptation’ mechanism for the knowledge modeled in the network. The final system also includes a sensor processing unit and a remote maintenance module system that provides an automated remote condition monitoring system, for both applications. Results indicate the feasibility of partial solutions in monitoring and diagnosis, though future enhancements are needed to compose a complete solution. This paper explains the characteristics of the Bayesian Network solution finally developed for high-speed machine tools, evaluate their strengths and weaknesses, and indicate the future enhancements.
Predictive maintenance strategies are very efficient in mechanical-failure modes, when failure probability increases with time, and one or more condition-monitoring techniques (vibrations, oil, thermography, sound,…) can predict the failure before breakage. However, use of these techniques have been always linked to a very limited kind of assets, where unavailability can lead to safety loss (Aerospace), or big economical losses (Energy).
One of the greatest problems in predictive maintenance is the need to handle great amounts of data and the cost in personnel training. To overcome this problem, several decision support systems are in development and research. However, the ill-structure of predictive data makes it difficult to automate the tasks, as there is normally uncertainty in the diagnosis process. Moreover, some diagnosis and monitoring models must be built straight from data as there is no experience available. Finally, there are cases where continuous adaptation of diagnostic and monitoring solutions is required, as knowledge in maintenance increases.
In order to cope with these problems, several algorithm paradigms coming from machine-learning field can be applied to a variety of situations, and with diverse requirements and problem characteristics (supervised, un-supervised, reinforcement learning, expert knowledge, data mining). Among these, there are some that have developed into robust tools that can be used for the modeling of several kind of classification problems, such as monitoring and diagnosis. Examples of successful paradigms are CBR systems, neural networks, induction trees & rules, and Bayesian networks.
A case study is presented, corresponding to EC partially supported MINICON project (minimum cost, minimum size, maximum benefit condition monitoring system), where a predictive maintenance solution for non-critical machinery (such as elevators and machine tools) was sought.
In the first place, this work exposes the necessities that can arise in the near future, within the application of intelligent systems for the maintenance of industrial assets. Thus, we have worked with maintenance that are currently covered by working inferential systems (i.e., expert systems) or are demanded within the industry. These problems have allowed to identify learning processes that can be necessary to complement and to support the inferential processes.
This paper also presents a description of available learning algorithms within Bayesian networks. The complete deployment and testing of learning capabilities is being made with HUGIN tool [5], which includes the necessary algorithms to implement the learning system. Learning can be considered as the true characteristic that makes a system to look like an intelligent device, automatically preventing the repetitive occurrence of silly errors. We have shown that Bayesian networks are algorithms worthy to be considered within the diagnosis area and failure detection. We include one prototype that shows how to adapt a system already built with partial information coming from incomplete sample sets.
There are, however, some issues that should devote additional work, beyond the scope of current activities. The gathering of appropriate data samples is a painstaking need, in order to properly ensure that any model works properly. Even though the data needs are low, since no training and testing sets are needed when experience is available, validation is still a necessary step. Validation has been performed using the data collected from several machines during the data analysis stage, and some more samples are coming from the pilot test.
We could also integrate synthetic data, taking into account controlled deviations from real samples, as well as theoretically known sample configurations. However, to become a true validation dataset, real samples with most of the diagnosis problems, and different ‘scenarios’ (machinery type, working conditions, etc.) should be taken. The collection of reliable data should be understood as of prime necessity in the machine tool sector, in order to validate any diagnostic alternative, and it can deserve a whole multi-national effort, bringing together past and future monitoring and diagnostic results.
Concerning the BN development, the work performed has not deployed the potential of the troubleshooting mechanisms, understood as utility or decision nodes in Bayesian networks, where normal diagnosis nodes (non-troubleshooting) are just referred to as belief nodes.
In conclusion, we can say that, there is a need for developing adaptive systems that can improve software support for maintenance decisions in many application fields besides machine tools and elevators. The work shown in this paper is a sound alternative to provide the necessary adaptation mechanisms.
