Abstract: Many works have been written on financial indicators to assess the use of a Maintenance Policy based on Total Productive Maintenance, while others have compared results showing the impact of criteria such as the Mean Time Between Failures. In our work, several indicators are proposed to enable maintenance managers to estimate the reliability level of a maintained process based on Total Productive Maintenance. The maintenance events used for analysis were collected by the operator of the machine (delegated maintenance) or by the maintenance team. These events were extracted from a database produced by a Computerized Maintenance Management Information System. No sensor was added to the system to evaluate its reliability level. The non-addition of a sensor makes it possible to minimize the financial costs during a first evaluation.

We present some indicators from different algorithms: Survival Laws, Remaining Useful Lifetime, the undesirable effect of certain actions on the process (Support Vector Machine), and a reliability indicator based on a Hidden Markov Model. We show that some statistical laws are not as effective as other methods. We show in this paper, that it's possible to estimate with a good quality, a strongly level of degradation (alarm) before the failure.

For example, a study of concrete cases improved the proposed approach on a discrete process (an alloy foundry). In this case, the maintenance manager can use different indicators on his computerized maintenance management information system to avoid, if possible, failure situations. We also show that a cumulative maintenance activity can sometimes degraded the process, like a cumulative drug can sometimes harm a sick patient.

Keywords: Total Productive Maintenance, diagnosis, prognosis, Remaining Useful Lifetime, Hidden Markov Model, decision support.

Connecter des objets numériques constitue un vecteur de transformation majeur de nature à bouleverser les modes de fonctionnement des sites de production, de gestion des immeubles, ou de maintenance, d’entretien de processus, de flottes de véhicules... De facto, un nombre quasi illimité de processus industriels, activités et services sont potentiellement concernés. En étant connectés, en devenant plus intelligents, les équipements, les processus industriels ou tertiaires génèrent un flot d’informations considérable. Rassemblées et exploitées, ces informations font apparaître de nouvelles opportunités et contraintes à appréhender.

L’intégrité des données sera le cœur de notre proposition. Elle consiste à maintenir la cohérence, la véracité et la fiabilité des données tout au long de leur cycle de vie (apparition, diffusion, traitement, utilisation, stockage ou destruction). 

Les données ne doivent pas être modifiées pendant leur transit. Il est donc indispensable de mettre en place un certain nombre de dispositifs afin de pouvoir détecter tous changements de contenu d’origine humaine ou liés aux dysfonctionnements du système de communication... Confidentiality, Integrity, and Availability (CIA triad).