We live in a fully connected world where, both in industries and for final users, the presence of “smart” devices has a heavy influence on lifestyle and production processes.
In particular, industrial systems and services are based on complex equipment connected to the network (Internet of Things), whose good operational status guarantees continuity and quality of production.
Any interruptions in the production processes, due to failures that determine the need for extraordinary maintenance, represent significant company costs.
It is here that the importance of predictive maintenance is manifested.
Traditional maintenance models
Today production plants’ maintenance is often preventive, based on planned interventions at precise time intervals, or reactive, based on extraordinary interventions when blocking malfunctions occur. These maintenance interventions are limited to adjusting the damage without storing the experience and information acquired to implement subsequent corrective actions in order to optimize the maintenance process.
Why predictive maintenance
Real-time IoT solutions allow companies to acquire large volumes of information on the operation of the equipment, enabling the execution of predictive models able of reliably foreseen failures and detecting degradation of production processes.
In this way it is possible to optimize maintenance activities, allowing a cheaper, more qualitative and more organizated business process.
Predictive maintenance (PdM) uses advanced big data analytics techniques for IoT sensor data processing
PdM enables decision-making and fault prediction during systems monitoring.
Revelis solutions also offer explanation functionality, that is, identification and interpretation of the root-cause of anomalies at the basis of the fault.
Predictive Models
By applying descriptive analysis techniques, machine learning and deep learning (neural networks), it’s possible to recognize anomalous behaviors (outlier detection) and / or to predict the residual useful life time of the equipment
Automated Reasoning
Thanks to reasoning engines based on Answer Set Programming it’s possible to enable the execution of logical rules for decision-support of end users
Explanation
Failures multidimensional analysis are used to highlight the malfunctions’ causes, allowing critical sub-systems repair with time savings and more efficiency
Big Data acquisition and storage
Sensor data acquisition requires robust and extensible IoT gateways, and Big Data storage is based on NoSql allowing time series management
Cloud Platform
Storage, processing and explanation features are made available in the cloud.
The platform could also be installed on-premises
Edge computing
Failure prediction models can be executed “on board”, minimizing reaction times and reducing the time and costs for transferring large amounts of data to the cloud
Predictive Maintenance advantages
A correct forecast of the failure-time or of the residual equipment life-time ensures maintenance process optimizations (i.e. for workforce scheduling, warehouse management, plant downtime minimization) and consequently in reducing direct and indirect costs
Real-time analysis
The product data management solutions allows continuous monitoring of civil plants and industrial plants. This monitoring occurs through operational dashboards and foressen fault-warning in (near) real-time
High-dimensionality control
Predictive maintenance combines approaches (occurrence of “static” conditions defined by domain experts) with failure prediction through inductive (machine learning and deep leanring) and deductive techniques (automatic reasoning)
New business models
Accurate prediction of failures, in addition to minimizing plant downtimes, allows the optimization of workforce planning, inventory management and supply chain management
Cost reduction
Cost reductions are very significative through the Predictive Maintenance and in industrial applications there is almost the 20% of saving on repairing and management costs
Predictive Maintenance Solutions key factors
Predictive maintenance is enabled by IoT sensors that continuously acquire data from the plant, processing it at the edge before transmitting to a cloud-based platform that enables real-time and historical analysis.
New technologies, such as distributed machine learning, advanced analysis and automated reasoning, play a vital role.
1
Big Data
Big Data is a large quantities of intermittent data streams (in real time or batch) from different data sources, mainly time series, which have different schemas and data structures
2
Artificial Intelligence Platform
Revelis solutions are based on Rialto™, an extensible and scalable artificial intelligence framework that can be easily integrated into customers’ operational contexts. Rialto™ allows the analysis/monitoring of large amounts of data, the forecasting of phenomena and the explanation of decision models
3
Machine/ Deep Learning
Supervised and unsupervised approaches, in addition to the use of Neural Networks (e.g. Recurrent Neural Networks)
4
Automated Reasoning
Answer Set Programming techniques, Ontologies and of Disjunctive Logic Programming
Predictive Maintenance Applications
Industrial plants
PdM could be used in production plants or transport infrastructures
Civil plants
PdM could be useful applied for lift monitoring and, more in general, in all electromechanical plants for domestic purposes
Wind farm, photovoltaic systems
By monitoring wind farm or photovoltaic systems, it’s possible to maximize electric energy production
