Risk-based Modelling for Wastewater Infrastructure Asset Management Zoran Kapelan University of Exeter International workshop on the science of asset management (FRMRC2) London 9 Dec 2011
Outline The framework Asset performance deterioration modelling Service impact modelling Intervention selection Summary
The Framework Optimal asset interventions determined by a trade-off between: Benefits arising from reduced risk of flooding / pollution as a result of intervention(s) and Associated costs of intervention(s) Risk estimated separately for each asset and for each failure mode In general case Risk = f (LF, CF) where: LF estimated from sewer performance deterioration models CF estimated using consequence / impact type models
Sewer Deterioration Sewers deteriorate with time Repair / refurbish / replace? Where? When? Traditionally decisions based on sewer condition - need to move toward performance based approach Repair Replace Condition and Performance Condition Performance Time
Sewer Performance and Serviceability WHY? HOW? WHERE? Sewer characteristics Sewer performance Sewer serviceability Material Shape/size Age Soil Type Collapses Blockages LF Flooding Pollution CF Condition Other Risk = f (LF, CF)
Sewer Performance Deterioration Modelling Aim: predict future sewer failure (blockage/collapse) rates from explanatory factors Data available: Incident records (observed Y) Sewer/other data (X) Issues: Unknown nature of relationship Lack of (reliable) data Different approaches exist Explanatory factors Sewer Performance Deterioration Model Blockage/ collapse rate
Evolutionary Polynomial Regression (EPR) Methodology and software tool for developing mathematical models from data Data driven approach
EPR Methodology Say we want to develop a mathematical model of polynomial structure that can predict the number of failures Y from a range of potential explanatory factors X: Fit the above model to the observed data assuming the following unknown parameters: Input variable s exponents ES jk m Polynomial coefficients a j ES j1... Y a a X X 0 j 1 k j 1 ES jk
EPR Methodology (cont.) Assume the following model fitting criteria: Maximise model fit: Minimise SSE Minimise model complexity: Minimise number of polynomial terms and/or Minimise number of input variables EPR solves the above problem as a multi-objective optimisation problem EPR outcome: A set of Pareto non-dominated models Model Complexity Model Fit (SSE)
EPR vs. Other Modelling Techniques
EPR Application Methodology developed as part of the UKWIR project Deterioration Rates of Sewers Findings published as UKWIR report No. 06/RG/05/15 Methodology used to develop PR09 models for all mains and sewers in a UK water company
Example: Deterioration Models Sewer Material Collapse Models for small diameter gravity sewers R 2 (%) N o Pipes Significant Explanatory Factors Asbestos Cement CR = a x A x L -1 x Gr 0.5 x Dia 0.5 x PCL 2 91 6,210 Age, Traffic & US Properties Brick CR = a x A 0.5 x PBL -1 x PCL x SHL01 2 + b x A 2 76 3,379 Dia, MinDepth & x Gr 2 x Dia 0.5 x PWC -1 x PCL x SHL01-0.5 Traffic Clay CR = a x TL 0.5 x PBL + b x NConn 0.5 x 61 471,982 Age, MinDepth & USNConn 0.5 + c x A 2 Upstream Props Concrete CR = a x Gr -2 x Dia -1 x MPC 2 x PBL 0.5 + b x A x PB -1 x PBL 63 142,573 Age, Dia & Upstream Properties GRP CR = a x A 2 x PB -0.5 x Dia x PBL 77 6939 Dia, MinDepth & Traffic
Bayesian-based Sewer Performance Modelling Based on Bayesian theory Zero-Inflated HNPP (mixture) model Differences when compared to EPR: Probabilistic approach (uncertainty bounds) Enables prediction of collapse / blockage rates at single pipe level Enables inclusion of additional information Provides framework for periodic model updating
Example of Bayesian-based Model Results
Asset Impact Modelling Integral part of service risk assessment - Risk = f (LF, CF) Estimated using different impact models, both physically and data based Typically requires large quantities of diverse type of data Some impacts difficult to determine
Example: Risk Assessment The impact of asset failure (CF) defined as a cost to service over a period of time (2007-32): Quantified service impact: number of measurable service impacts occurring following asset failure Service impact unit cost: the cost incurred from the occurrence of a service failure
Service Impacts Quantified Service Impacts Asset Groups Failure Modes Flooding Internal Flooding External Curtilage Flooding - External Highway Flooding - External Other Pollution Categories 1 to 4 Traffic Disruption Health & Safety Surcharge Blockage Collapse Large diameter sewers Burst Small diameter sewers Ex-section 24 sewers Rising mains Vacuum sewers CSO structures Sewage pumping station
Service Impact Modelling Service Impact Modelling Asset Group Failure Mode Modelling Source > Pathway > Receptor Historic Impact Large diameter gravity sewers EPR Expert View Historic Performance Small diameter gravity sewers Ex-Section 24 sewers Rising Mains Vacuum sewers CSO structures Sewage Pumping station
Source > Pathway > Receptor Impact Modelling Flooding modelling: Source: FastNett, static and rising mains models Pathway: FloodRiskMapper Receptor: Internal/external flooding, road traffic disruption, h&s Pollution modelling: BOD dilution factor Requires large quantities of data but can be easily applied to large numbers of assets Allows the service impact to be determined for those assets which have not failed historically
Surface Grid Mapping
Examples of Flood Pathways
Historic Service Impact Modelling Work order data is analysed at an asset group or regional level to determine historic levels of service and the proportions of service impacts associated with an asset failure The service impact characteristics are assumed to not change with time This approach has been used for assets with: where the asset is not data rich historic low levels of service impact
Asset Intervention Selection Need to decide what to do, where and when over some long-term planning horizon not easy! Optimal interventions determined by addressing the trade-off between: Potential service improvement (i.e. future risk reduction) Associated cost required to deliver the improved service Improved service can be defined using a set of different criteria Ranking vs optimisation Cost Benefit
Summary Need to move from condition to performance / serviceability based asset management that is: Forward-looking Based on predictive models Risk-based Proactive Integrated Leads to selection of optimal interventions with respect to multiple criteria
Thank you! Questions? z.kapelan@exeter.ac.uk www.ex.ac.uk/cws