Periodic Reporting for period 2 - INFRALERT (LINEAR INFRASTRUCTURE EFFICIENCY IMPROVEMENT BY AUTOMATED LEARNING AND OPTIMISED PREDICTIVE MAINTENANCE TECHNIQUES)
Reporting period: 2016-11-01 to 2018-04-30
- The Data Farm allows taking advantage of useful asset data that was not considered before by system operators.
- Asset forecasting is paramount for advanced maintenance planning and decision making, but nowcasting models are also relevant to derive asset condition to evaluate the effect of maintenance actions.
- Alert systems guarantee safety, integrity and performance of linear assets, avoiding additional corrective actions and unnecessary preventive ones.
- The RAMS methodology presented demonstrates the application at system level based on the intervention-component approach, what allows calculating LCC using cost databases.
- Stochastic and dynamic optimisation allows handling uncertainties occurring in the planning process and designing advanced decision support systems.
- Cloud-based software and a distributed middleware architecture provides open and standard interfaces to support data exchange with external systems and scalability.
In WP2 an integrated data storage has been created through the implementation of an innovative Data Farm able to store data representing linear infrastructures and efficiently manage Big Data coming from the field. Techniques related to data uploading and data retrieving in a cloud computing environment have been integrated. Particular attention has been paid to data localisation procedures.
WP3 started with the development of a framework for the hierarchy of condition information and selection of relevant indicators. Methodologies for asset condition nowcasting and forecasting were described, including several types of models for the asset deterioration modelling: symbolic, data-driven, and physics-based. Work on hybrid methodologies and uncertainty modelling for the asset condition toolkit was performed in the second period.
Work done in WP4 covers the development of several machine learning procedures to generate maintenance alerts from infrastructure conditions. An automatic learning methodology for estimating alerts, intervention types and the probability of occurrence based on work-orders, asset features and measurement auscultations from historical interventions was implemented. The toolkit developed includes self-learning rules for automatic learning from false predictions.
WP5 focussed on the development of models and a methodology to calculate RAMS and LCC parameters from data at component and system level for rail and road assets. The implementation of algorithms to calculate RAMS&LCC in real environment for switch and crossings (rail) and pavement system (road) was carried out.
In WP6, for the planning on tactical and operational level mathematical models for maintenance decision support have been developed and implemented as software toolkits. In particular, uncertainties resulting from asset conditions are considered during the optimisation. For the strategic planning a simulation framework to consider long-term strategies was compiled using the developed toolkits.
In order to fulfil defined system requirements as much as possible the eIMS developed in WP7 is composed of easily scalable multi-layers development and system architectural environments. eIMS is cloud-based, provides intelligent mapping frameworks independent from the used database. The Integration Gateway provides open interfaces to support the data exchange with external systems to avoid replication of existing data.
WP8 involved the integration of all necessary data, from measurements to historical maintenance interventions including collecting and updating data. It then allowed testing the eIMS prototype for both case studies encompassing all stages: Data Management, Data Analytics and Decision Support. Besides this technical validation of the eIMS implementation, WP8 addressed the impact analysis of INFRALERT by comparing the eIMS outputs to the current practice, using the defined evaluation framework and KPIs. It also allowed assessing the applicability in operational practice from a user perspective in qualitative terms.
In WP9, dissemination was performed through different activities (e.g. 11 publications, 29 press releases, 2 project-focused workshops, 3 EAB meetings, attendance at 23 conferences, exhibitions and similar events) using different material (e.g. 2 leaflets, 6 posters, 3 newsletters) and channels (e.g. social media, website). The exploitation strategy is focussed on the commercialisation of the developed eIMS as well as the individual toolkits.
- New global ontology behind the topological description of linear infrastructures: general concepts of nodes and edges for rails and roads representations.
- Hybrid Data Farm as innovative concept for surpassing the barrier of choosing between two alternative types of databases (relational vs. non-relational).
- Statistical and visualization methods for linear asset condition assessment using predefined limits and hybrid modelling framework for road and railway infrastructure.
- Models for asset condition data pre-processing, infrastructure degradation and maintenance effectiveness for both short-term (operational) and long-term (tactical) assessment as well as asset condition prediction.
- Regarding alert generation, the new methodology exploits historical maintenance database to characterise asset condition based on explanatory features, listing and ranking the severity level of assets without previous predefined thresholds (machine learning).
- New methodology to calculate RAMS&LCC parameters at system level using the component-intervention approach and data processing algorithms for work-orders. The main innovation is the application for pavement road system.
- Handling uncertainties occurring in the planning process of linear infrastructures. The novel approach includes stochastic and dynamic optimisation to generate robust maintenance solutions.
- eIMS architecture to integrate different toolkits with an integration gateway for external systems, assuring the interoperability.
Summary of socio-economic impact related to current practices:
- Rail networks: increased reliability (57%) and availability (6%), energy and cost savings (53%), improvement on traffic disruption (6%), safety (65%) and comfort (82%)
- Road pavement system: reliability (+59%), savings in maintenance costs (9%) and energy/CO2 emission (10%), availability (+43%), improvement on traffic disruption (1%), safety and comfort (+26%)