The National Academies Keck Center, Washington DC March 10, 2011 1 2 Outside stakeholders Internal participants Senior management Executives • Districts • Modal units • Engineering disciplines • Planning • Design • Maintenance The Public Elected Officials Governor • legislature • county commissioners • city council Appointed Oversight Transportation commission • MPO board Planning and support Asset management leadership Asset management director • Bridge management engineer • Pavement management engineer Funding Bodies FHWA • FTA Interest Groups Highway users • homeowners associations • business groups • constituency groups Budget/finance • Program management • Strategic planning • Public information • Information technology Maintenance staff Maintenance engineers/ managers • Facility managers • Maintenance crew leaders • Emergency response Engineering staff Project engineers and managers • pavement surveys • materials/research • bridge design/rating • bridge inspection Roles in asset management 3 Senior management – top-down vision Oversight bodies – make service life tangible Asset managers – decision outcome measure Practitioners – Learn how to compute and present life expectancy Engineers and planners – Learn how to use life expectancy in design and planning System designers – How to build life expectancy into software and tools Researchers – Improve state of the practice 4 1 Define the scope Set goals and objectives Identify desired applications Identify network of interest Identify asset types Assess gaps and readiness Planning 2 How to use this guide Plan for implementation Document business processes Plan the change strategy List desired reports and tools Define work plan, resources Set quality metrics, milestones 3 How to plan life expectancy models Establish the framework Define performance measures Conceptualize the analysis Determine data requirements Mock up tools and reports Gain buy-in, build expectations How to design life expectancy models 4 Develop foundation tools Prototype lifespan calculations Evaluate prototype results Refine computations Implement foundation tools Document methods and tools How to compute life expectancy models Development 5 Develop applications Prepare user group Prototype applications Pilot test and evaluate tools Refine and roll out Document tools, procedures How to apply life expectancy models 6 Evaluate and refine Assess quality, sensitivity Improve model realism How to improve life expectancy models 7 Prolong implementation Measure, promote success Add to management systems How to perpetuate life expectancy models 5 Justify maintenance funding Plan timing and scope of actions Plan staffing and equipment Set inventory levels Evaluate new materials, methods Reduce workzone frequency Improve consistency of reports Optimize cash flow Build credibility 6 Life expectancy if no maintenance Life expectancy under a proposed maint policy Life extension effects of preservation actions Compare preservation alternatives Optimal replacement interval Optimal preventive maintenance interval Optimal expenditure on periodic maint Scope and timing to maximize life extension 7 Compare design alternatives using life cycle cost Price point where a new material is attractive Coordinate replacement of multiple assets Plan corridor work zones and traffic control Multi-objective prioritization Funding allocation and effect of budget cuts Select treatment application policies Establish research priorities 8 Start small, build incrementally Prototype or proof-ofconcept Pilot test or experimental application Statewide limited rollout Expansion to agency-wide and to partner agencies 9 Asset management maturity scale Maturity Level Generalized Description Initial No effective support from strategy, processes, or tools. There can be lack of motivation to change improve. Awakening Recognition of a need, and basic data collection. There is often reliance on heroic effort of individuals. Structured Shared understanding, motivation, and coordination. Development of processes and tools. Proficient Expectations and accountability drawn from asset management strategy, processes, and tools. Best Practice Asset management strategies, processes, and tools are routinely evaluated and improved. 10 Part A. Policy Guidance. How does policy guidance benefit from improved asset management practice? Policy guidance benefitting from good asset management practice Strong framework for performance-based resource allocation Proactive role in policy formulation Part B. Planning and Programming Do Resource allocation decisions reflect good practice in asset management? Consideration of alternatives in planning and programming Performance-based planning and a clear linkage among policy, planning and programming Performance-based programming processes Part C. Program Delivery Do program delivery processes reflect industry good practices? Consideration of alternative project delivery mechanisms Effective program management Cost tracking and estimating Part D. Information and Analysis Do information resources effectively support asset management policies and decisions? Effective and efficient data collection Information integration and access Use of decision-support tools System monitoring and feedback 11 Get ducks in a row Policies in place Procedures defined Ability to deliver planned actions Availability of data Decide how far to reach in next 2-3 years Visualize agency capabilities at the end Create implementation plan How to get from here to there 12 13 14 Asset management tools, such as life expectancy analysis, are built in order to improve the way your agency does business. Organizational change can be beneficial, and can be scary. You need a vision and a strategy in order to be successful. 15 Credible long-term view of asset performance Accountability (benefits and fears) Tangible levels of service Understanding of deterioration and growth Optimal preservation Improved competitiveness for funding Constructive political relationships Be ready to follow through to win these benefits 16 Why? Ensure the tools are relevant Understand how they will be used Build the right tools for the job Select appropriate methods Help others understand Gain buy-in Gather data: Inventory • Geodata Condition • Traffic Risk • Safety Inspect reports Assess data quality Set minimum tolerable performance Prioritize for further development Develop deterioration models Identify assets needing work Needs Develop life expectancy models Design rehabilitation actions Develop corridor plans Evaluate market conditions Find economies of scale Corridor plans Develop cost models Develop work packages as projects Develop effectiveness models Project plans Designs Develop budget constraints Evaluate equity Evaluate fiscal uncertainty Negotiate with funding bodies Prioritize and schedule STIP Monitor performance Select rehabilitation actions Develop performance targets Plan for delivery Lettings Annual 17Reports Convince staff of the need and benefit of the change and the tools Create a change leadership coalition Develop a vision of the end result Communicate the vision regularly Take actions consistent with the vision Make sure staff are involved and empowered Show short-term successes Keep the focus on the change effort Anchor new approaches into the culture 18 1. 2. 3. 4. 5. Data acquisition and management Plan foundation analysis methods List/describe applications and reports Write a work plan Set quality metrics and milestones 19 Geo-referencing Traffic counts Crashes Asset inventory Asset condition Asset vulnerability Climate Soils NOAA Climate Divisions 20 Considerations: Purpose of the tools Types of assets to be addressed Performance measures Define end-of-life Define intervention possibilities Account for uncertainty Analysis level: • Network level – Life expectancy of families of assets based on general characteristics • Project level – Life expectancy of a single asset based on age, condition, and asset characteristics 21 Probability Considerations: Subject matter Filtering Aggregation Sorting Graphics 1.0 Average 0.9 0.8 0.7 Cumulative 0.6 This year 0.5 0.4 0.3 0.2 0.1 0.0 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 22 60 Age Task 1. Define scope of the analysis. Task 2. Develop implementation plan. Task 3. Define performance metrics and analysis concepts, including data requirements and mock-ups. Task 4. Develop foundation tools and models. Task 5. Build applications, possibly through a series of prototypes. Task 6. Ensure long-term support. Evaluate usage of the product and make improvements. 23 24 25 Value Straight-line depreciation Interval replacement Value Age Premature failure Age 26 Performance, condition, or value Performance, condition, or value Decisionsensitive Deterioration model End-of-life threshold Age End-of-life threshold Age 27 Performance or condition Replacement: Extended life = 10 years, Cost = $100,000 Repair: Extended life = 4 years, Cost = $50,000 End-of-life threshold You are here (end of 10 year life) Age 28 Life expectancy depends on how you define the end-of-life. Agencies may often have a degree of control over life expectancy. Lifespan can often be managed to maximize agency objectives or minimize life cycle costs. 29 Performance, condition, or value Sudden failure Performance, condition, or value Age Standard Obsolescence due to raised standard Age 30 Performance, condition, or value End-of-life defined by age Performance unknown, not measured, or doesn't matter Remaining capacity, stock, or value End-of-life based on utilization Age Consumption or utilization rate Age 31 Probabilistic end-of-life Probability of failure Optimal replacement interval Pavement condition Median time to fail End-of-life from terminal criteria Roughness Cracking Age First to fail End-of-life threshold Age 32 Normal deck life expectancy 20 years Bridge condition Normal substructure life expectancy 50 years Substructure rehab adds 10 more years, allows full utilization of the third deck End-of-life threshold Age 33 Bridge condition, performance Plan for two deck rehab projects to extend deck life until ready for replacement Traffic forecast calls for unacceptable level of service after 30 years End-of-life threshold Age 34 Remaining service life Condition Current condition Life extension End-of-life threshold Age 35 Probability of failure 20% will have failed by 10 years Median time to fail (life expectancy) = 12 years Program period ends at 10 years Age 36 Techniques are related to deterioration modeling, but usually simpler. Select a method based on the kind of data available, the needs of the application, and the importance of uncertainty 37 Performance, condition, or value Continuous Deterministic Performance, condition, or value Age Performance, condition, or value Continuous Probabilistic Age Discrete Deterministic Age Performance, condition, or value Discrete Probabilistic Age 38 Visual inspection (100% sample) 10% sample of road segments Automated data collection 39 Digital dashboards 40 Using Excel for report mock-ups 41 Using Excel for application development 42 43 44 What to Model What to Model Influence of Framework Model Selection Model Selection Selection Criteria Data Availability Estimation Techniques Nature of Prediction and Outcome Estimation Techniques Regression Conclusion Survival Models Markov Chains 45 45 What to Model Model Selection Estimation Techniques Conclusion End-of-Life can be taken as the time until ▪ Functional Obsolescence ▪ Changes in standards ▪ Changes in functional requirements ▪ Structural Deficiency ▪ Deterioration ▪ Extreme events If modeled separately – Min. life assumed If combined – Direct prediction of life 46 46 What to Model Model Selection Estimation Techniques Two general approaches Interval-based ▪ Predict time until end-of-life event occurs ▪ Directly predict life based on historical replacement intervals Construction, X Reconstruction, Y Service Life Conclusion Year Year TX Year TY 47 47 What to Model Model Selection Estimation Techniques Conclusion Two general approaches Condition-based ▪ Predict condition or measure of performance as a function of time ▪ Predict asset value as a function of time Performance, condition, or value Deterioration model End-of-life threshold Performance, condition, or value Decisionsensitive End-of-life threshold Age 48 Age 48 What to Model Model Selection Estimation Techniques Conclusion General Criteria Transparent ▪ Staff Knowledge ▪ Able to Replicate and Revise Applicable ▪ Data Availability ▪ Widespread Use of Results Focused ▪ Prioritize on Predicting Life ▪ Not necessarily Deterioration-based 49 49 What to Model Model Selection Estimation Techniques Model Selection depends on Data Availability ▪ Historical Service Life ▪ Dominating end-of-life condition preferred ▪ Condition Data by Age ▪ Archived Data Preferred Conclusion 50 50 What to Model Model Selection Estimation Techniques Model Selection depends on Nature of Dependent Variable ▪ Continuous Variable ▪ Time until rationale event occurs ▪ Performance Measures (e.g., IRI, Rutting, NBI Sufficiency Rating) ▪ Discrete Variable ▪ Performance Measures (e.g. NBI element Condition Rating, PSI) Conclusion 51 51 What to Model Model Selection Estimation Techniques Model Selection depends on Nature of End Result ▪ Deterministic Performance, condition, or value Continuous Performance, condition, or value Discrete Age ▪ Probabilistic Performance, condition, or value Continuous Age Performance, condition, or value Discrete Conclusion Age Age 52 52 What to Model Model Selection Probabilistic estimates can be represented by Density functions 0.14 0.12 Estimation Techniques Probability 0.10 Probability Density Function 0.08 0.06 Median 0.04 Confidence Interval 0.02 Conclusion 0.00 30 35 40 45 50 Service Life in years 55 60 53 53 Model Selection Estimation Techniques Conclusion Probabilistic estimates can be represented by Survival or Cumulative functions ▪ Survival Prob. = 1 - Cum. Prob. Probability of Passing What to Model 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Survival Function Median Confidence Interval 30 35 40 45 50 Service Life in years 55 60 54 54 What to Model Model Selection Estimation Techniques Conclusion Basic Techniques Deterministic ▪ Regression (Continuous Data) Probabilistic ▪ Simple Average (Continuous Data) ▪ Survival Models (Continuous Data) ▪ Markov Chains (Discrete Data) Alternatively, may be forced to rely on published life expectancy values or expert opinion 55 55 What to Model Data requirements Requires historical replacement data Does not require explanatory factors Model Selection Method Fits distributions to groups of assets based on average and standard deviation of data Estimation Techniques Average age at replacem ent a is culvert age, N is number of culverts Population standard deviation (use if list is w hole population) Conclusion Sam ple standard deviation (use if list is a random sample) s is an estimate of σ s N 1 a N a a a 1 N ai a N 1 i 1 1 N i 1 N i 1 i i 2 2 56 56 What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 57 57 What to Model Model Selection Estimation Techniques Data requirements Requires ▪ Historical replacement data or Continuous performance/condition data & age ▪ Set of independent, explanatory factors Method Predicts dependent variable as a function of explanatory factors ▪ E.g., predict life as a function of traffic volume, maintenance history, material type, climate conditions, etc. Life Prediction Conclusion t b1 X 1 b 2 X 2 b n X n 58 58 What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 59 59 What to Model Model Selection Estimation Techniques Data requirements Historical replacement data or Time until end-of-life criteria reached Set of independent, explanatory variables Method Predicts survival curve (% assets passing beyond point in time) as a function of explanatory variables No assumption of statistical distribution Median life = 50% survival probability Probability of Passing Conclusion y 1 g exp 1 . 0 g / exp b 1 X 1 b2 X 2 bn X n 60 60 What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 61 61 What to Model Model Selection Estimation Techniques Data requirements Historical replacement data or Time until end-of-life criteria reached Method Predicts survival curve (% assets passing beyond point in time) Probabilities governed by Weibull distribution (or Markov/Exponential model if shape parameter = 1) Median life = 50% survival probability Probability of Passing Conclusion y1g exp 1.0 g / 62 62 What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 63 63 What to Model Model Selection Estimation Techniques Improves upon Quick-and-Simple Weibull technique by adjusting predictions to a set of independent, life expectancy factors Probability of Passing y1g exp 1.0 g / expb1 X1 b2 X 2 bn X n Example Demonstration Conclusion 64 64 What to Model Model Selection Common technique for predicting the probability of being in any discrete condition state at any point in time Pii ≡ Probability of staying in same condition state i after unit time Pij ≡ Probability of transitioning from state i to a worse condition state j after unit time Estimation Techniques Good Fair Poor Conclusion 65 65 What to Model Probabilities represented in matrix form PFF=93.2 PGG=95.3 PFP=3.9 PGF=4.6 Model Selection Good PPP=100.0 Fair Poor PGP=0.1 Estimation Techniques Conclusion Markov transition probability matrix State State probability in one year Today Good Fair Poor Good 95.3 4.6 0.1 Fair 0 93.2 3.9 Poor 0 0 100.0 66 66 What to Model Data Requirements Pairs of inspection Data with Discrete Condition Rating Model Selection Method Estimate transition probability between 2 states: ‘failed’ Estimation Techniques and ‘not failed’ Compares % assets in each condition state from one year to the next Median Life taken as Conclusion 67 67 Quick-and-Dirty Markov Chain What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 68 68 What to Model Model Selection Estimation Techniques Similar to Quick-and-Dirty but now analyzes multiple (>2) states Data Requirements Transition probabilities by way of expert opinion, observed frequency, optimization, one-step process, etc. Method Probabilistic estimate of condition states by age Median Life = 50% assets in threshold state Probability of state k next year: y k x j p jk for all k j Conclusion j is the condition state this year and x is the fraction in state j p is the transition probability from j to k 69 69 Markov Chain What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 70 70 What to Model Data Requirements Pairs of inspection Data with Discrete Condition Rating Model Selection Estimation Techniques Method Predicts transition probabilities by comparing % assets in a condition state at the end of the year to that at the beginning of the year Assumes condition state never drops more than one step per year Life prediction same as previous example Conclusion 71 71 One-Step Process What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 72 72 What to Model Data Requirements Transition probabilities by way of expert opinion, observed frequency, optimization, one-step process, etc. Model Selection Method Predict age as a function of condition Calculate condition index weighted by time spent in Estimation Techniques Conclusion condition state or lower state Approach converts a Markov model into a Weibull model log(0.5) tj log( p jj ) log lnCI g 10^ g is equivalent age CI is condition index 73 73 Equivalent Age Markov What to Model Model Selection Example Demonstration Estimation Techniques Conclusion 74 74 What to Model ▪ End-of-Life Definition(s) Needed ▪ Interval- or Condition-based Approaches Model Selection Estimation Techniques ▪ Selected Models should be transparent, applicable, and focused ▪ Selection influenced by nature of dependent variable and estimate ▪ Basic modeling techniques include Conclusion ▪ Regression ▪ Survival Models ▪ Markov Chains 75 76 77 Presentation Outline APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users Conclusion • Life Expectancy Estimates from Deterioration Model • Additional Building Blocks for Life Expectancy Application • Example Applications • User Groups • Conclusion 78 Deterioration Model APPLYING THE MODELS Deterioration Model • Life expectancy estimates -- easily derived from deterioration models • Additional tools are developed on top of life expectancy estimate to help management decision making process Building Blocks Example Role of Users Conclusion 79 Additional Building Blocks for Life Expectancy Application APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users • Techniques of life expectancy analysis open the door for many useful applications to support TAM decision making, but few more building blocks are required: – – – – Equivalent age Life extension benefits of actions Remaining service life Life cycle cost models Conclusion 80 Equivalent age APPLYING THE MODELS Deterioration Model Building Blocks Example • Deterioration models often use age of an asset to forecast its condition • However, many applications require finding out ‘equivalent age’ from known condition of an asset Role of Users Conclusion 81 Life Extension Benefits of Actions APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users • Effect of repair & rehabilitation actions is expressed as an improvement in condition • Once the improved condition is forecast, we can find equivalent age, before and after the action • The difference in age is one way of expressing the benefit of the action Condition Change in equivalent age = Life extension benefit Conclusion Life extension action improves condition Age 82 Remaining Service Life APPLYING THE MODELS Deterioration Model Building Blocks Example • Computed by subtracting actual age of an asset from its life expectancy (provided no repair was done) • If an asset has been repaired, it is more accurate to use a condition-based approach (i.e., taking advantage of deterioration and equivalent age models) Unknown past work Remaining life Condition Role of Users Conclusion Current condition End-of-life threshold Age •Current condition of the asset can be converted to its equivalent age, which is then subtracted from life expectancy to estimate remaining service life 83 Life Cycle Cost Models APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users Conclusion • Life cycle cost models, combined with life expectancy and deterioration models, may be used in numerous useful applications to support TAM decision making • Few concepts associated with life cycle cost models – – – – Time value of money Benefit/cost ratio Comparing alternatives using Net Present Value (NPV) Comparing alternatives using Equivalent Uniform Annual Cost (EUAC) – Comparing alternatives using Present Worth at Perpetuity – Comparing alternatives using Internal Rate of Return (IRR) 84 Example Applications APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users Conclusion Many useful asset management applications can be created using the building blocks discussed – – – – – – – – Routine preventive maintenance Optimal replacement interval Comparing and optimizing design alternatives Comparing and optimizing life extension alternatives Pricing design and preservation alternatives Synchronizing replacements Effect of funding constraints Value of life expectancy information 85 Routine Preventive Maintenance APPLYING THE MODELS Deterioration Model • An example of comparing a preventive maintenance scenario against do-nothing scenario Year Building Blocks Example Role of Users Conclusion 1 ... 4 ... 8 ... 12 ... 16 ... 20 ... 24 Cost per lane-mile by strategy Do-Nothing Routine Preventive Maintenance $400 $400 $400 $400 $400 $30,000 $30,000 86 Routine Preventive Maintenance (contd.) APPLYING THE MODELS Deterioration Model Building Blocks • Let us assume, interest rate = 4% • The EUAC of the two alternatives can be compared as follows: = $9,083/lane-mile Example Role of Users = $596/lane-mile Conclusion = $768/lane-mile • In this example, the agency could reduce annual costs by $172 per lane-mile if routine preventive maintenance is completed 87 Optimal Replacement Interval APPLYING THE MODELS Deterioration Model Building Blocks • Assets may have a number of service life alternatives, depending on different strategies for maintenance and life extension • Optimal service life would be the life cycle activity profile that can be sustained at minimum life cycle cost • Here is an example of comparing several alternative profiles Example Role of Users Conclusion Replacement Cost Rehabilitation Cost Annual Maintenance Cost Estimated service life (N) Rehabilitation years Interest rate Compounded Life Cycle Cost Present Worth at Perpetuity Option 1 600 200 Option 2 600 200 Option 3 600 200 Option 4 600 200 5 50 25 40 5 60 25 45 0.05 0.05 5 70 25 45 55 0.05 5 80 20 40 60 0.05 $7884 $12727 $21146 $35411 $753 $720 $719 $729 88 APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users Present worth at Perpetuity ($1000) Optimal Replacement Interval (contd.) 760 Option 1 750 740 Option 4 730 720 Option 2 Option 3 710 700 40 50 60 70 Replacement cycle (year) 80 90 • Plot suggests that options 2& 3 are preferred and the optimal interval for replacement is between 60-70 years Conclusion 89 Comparing/optimizing Design Alternatives APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users • Comparing two products or methods that have different costs, different life expectancies, and different life extension possibilities • Here is an example, deciding on whether to apply coating to a pipe culvert – a non-coated culvert, expected to survive 50 years with a construction cost of $1000, and a coated culvert, expected to survive 56 years with a construction cost of $1200 Conclusion Therefore, the coated design option is preferred 90 Pricing Design and Preservation Alternatives APPLYING THE MODELS Deterioration Model Building Blocks • Many agencies have active research programs to develop new and improved maintenance materials and techniques • But, how cheap does it need to be before it’s worth using? Example Role of Users Conclusion • The methods of life expectancy analysis can often play a part in this evaluation – Example: To assess feasibility of switching from traditional carbon steel reinforcement bars to solid stainless steel reinforcement bars 91 Pricing Design and Preservation Alternatives (contd.) APPLYING THE MODELS ILLUSTRATION: Material for bridge deck reinforcement At what price ratio is stainless steel (SS) more cost-effective than traditional steel (TS)? Building Blocks Example Role of Users Conclusion Answer: depends on service life of each alternative FHWA Laboratory and field simulations: SS – 100 years (no deck replacement) TS – 70 years (1 deck replacement, 2 deck rehabs) Ratio of EUAC for Stainless Steel to Traditional steel Deterioration Model 1.1 1 0.9 0.8 Threshold Ratio Current Ratio 0.7 0.6 0 2 4 6 8 10 Ratio of Stainless Steel Price to Traditional Steel Price Source: Cope et al. (2011) Stainless Steel is MORE cost-effective Stainless Steel is LESS cost-effective 92 Effect of Funding Constraints APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users Conclusion • Decision support tools based on life expectancy and life cycle cost can help an agency to do more with less – Example: an agency calculated utility of a set of projects with respect to life expectancy, deterioration, life cycle cost, and estimated project cost. Let budget be $2.75M Activity Utility Cost Bridge A replacement 100 $2400k Bridge B rehabilitation 75 $250k Box Culvert A replacement 55 $100k Pipe Culvert A replacement 35 $5k Bridge C deck patching 32 $20k 93 Effect of Funding Constraints (contd.) APPLYING THE MODELS • Optimization techniques can be applied to select a set of projects (Solver option in Excel may be used) Deterioration Model Building Blocks Example Role of Users Conclusion Activity Utility Cost Bridge A replacement 100 $2400k Bridge B rehabilitation 75 $250k Box Culvert A replacement 55 $100k Pipe Culvert A replacement 35 $5k Bridge C deck patching 32 $20k Optimal solution: Total utility 242 at a cost $2.675M; remaining $75k to be carried over 94 Role of User Groups APPLYING THE MODELS Deterioration Model Building Blocks Example • One of the best ways to create involvement and buy-in is to form a user group for the applications that are to be developed • A user group should consist of people who will be hands-on users of the applications, as well as people who may receive and act on the information Senior management Role of User Groups Outside stakeholders Asset management leadership President Conclusion Steering/leadership committee Subcommittee Subcommittee Subcommittee 95 Role of User Groups (contd.) APPLYING THE MODELS Deterioration Model Building Blocks Example Role of User Groups Conclusion • The user group’s tasks include planning, development, & production of different applications • Often the user group will be large and may expand over time to include all hands-on users and many indirect users of the applications • Once the group reaches sufficient size, it should create sub-groups to whom it delegates many of the tasks above 96 Conclusion APPLYING THE MODELS Deterioration Model Building Blocks Example Role of Users Conclusion • An agency may launch a big system development effort to implement various applications of lifecycle estimations • Alternatively, it can select relatively small subset of applications at first (often just one), and develop working prototype – The prototype addresses core functions, from data collection to analysis & reports – Should gradually expand to cover more applications and to add more features – Should identify data gaps, procedures and standards that are required, in the context of a working application 97 Conclusion APPLYING THE MODELS Deterioration Model – It gives users more day-to-day control and involves them more deeply in the creation of the tools they will use, thus helps avoiding “not invented here” syndrome Building Blocks Example Role of Users Conclusion 98 99 100 Rationale Causes Sensitivity Risk Rationale for Incorporating Uncertainty Causes of Uncertainty Sensitivity Analysis Risk Analysis Uncertain Inputs Uncertain Outputs Conclusion 101 101 Rationale Life expectancy estimates affect business processes Causes Network Planning Preservation Policy Finance Budgeting Sensitivity Human Resources Corridor Development Programming Risk Design Conclusion Maintenance Project Development Preservation Planning Life Expectancy Analysis Information Technology Research Data collection but asset life is inherently uncertain… 102102 Rationale Causes Sensitivity Uncertainty result of random Random Process Example Structural Response Actual strength unknown due to material imperfections Loadings Uncertainty surrounding future traffic levels and % trucks Site Conditions Uncertain soil properties. future climate conditions, or random extreme weather events Human Influence Unknown construction and/or inspection rating quality Externalities Unforeseen development of new technologies or standards Risk Conclusion 103 103 Rationale Methods to quantify uncertainty Characteristic Causes Sensitivity Sensitivity Analysis Risk Analysis Nature of Outcome Deterministic Probabilistic Assesses how Outcome varies due to... Random Changes Unit Changes Risk Conclusion - Both can be used to produce ranges of life estimates - Risk analysis additionally describes the likelihood of life estimates 104104 Rationale Causes Sensitivity Benefits Identify most influential factors Guide design selections Assess potential life extensions Plan for mitigation Risk Conclusion 105105 Rationale Causes Analysis varies by model selection For models without explanatory variables, can assess how life prediction varies for different groupings of assets Sensitivity Risk Conclusion 106106 Rationale Causes Analysis varies by model selection For models with explanatory variables, can assess how life prediction changes when vary factors over a range of values Sensitivity Risk Conclusion 107 107 Rationale Causes Sensitivity Analysis varies by model type For Ordinary Regression ▪ Unit Δ in factor = β Δ in life prediction For Cox Regression models ▪ Unit Δ in factor = exp(β) % Δ in Hazard Ratio Risk For Weibull Regression models Conclusion ▪ Unit Δ in factor = exp(β) % Δ in Average Life where β represents the parameter estimate 108108 Rationale Tornado Diagram Representation Δ in Life Predictions Factor 1 Factor 2 Causes . Sensitivity Increasing Influence on Life . . . Risk . . Factor n Conclusion Increase in Factor leads to a Decrease in Life Increase in Factor leads to an Increase . in Life 109109 Rationale Causes Example Demonstration Sensitivity Risk Conclusion 110 110 Rationale Causes Sensitivity Risk To mitigate uncertainty, probabilistic techniques emphasized Describe likelihood of life expectancy and related business processes Ranges of life produced by level of confidence (μ point estimate) Probability of failure Median time to fail (life expectancy) = 12 years 20% will have failed by 10 years Conclusion Program period ends at 10 years Age 111 111 Rationale Describe Likelihood and Consequence of Risk Causes Sensitivity Risk Risk Identification Risk Assessment Quantify Likelihood and Consequence Risk Management Decide on Mitigation Strategy Conclusion Risk Monitoring Monitor Effectiveness of Strategy 112 112 Rationale Risk Assessment Process Causes X Sensitivity Risk Y Step 1: Quantify uncertainty surrounding life expectancy factors (e.g., climate conditions, traffic loading) using probability distributions Conclusion Z <Van Dorp, 2009 – GWU> O 113 113 Rationale Risk Assessment Process Causes X Sensitivity Risk Y Conclusion Z Step 2: Randomly sample input distributions and calculate life using the calibrated model <Van Dorp, 2009 – GWU> O 114 114 Rationale Causes Risk Assessment Process X Sensitivity Y Risk Conclusion Z Step 3: Assess the distribution of life estimates <Van Dorp, 2009O – GWU> 115 115 Rationale Causes Suppose an agency is interested in the risk of potential climate change on business processes Annual Costs Sensitivity Climate Risk Conclusion Service Life Propagating Uncertainty Budget Needs Project Utility 116 116 Rationale Causes Assume 30 year old bridge asset with following characteristics: Normal annual temperature (°F) = 49 Normal annual precipitation (in.) = 43 Sensitivity Part of NHS system Non-Corrosive Soil Risk Steel, girder bridge 50 feet long Wearing Surface Applied Conclusion $50k Replacement Cost 4% Interest Rate 117 117 Rationale Causes Assess how life changes due to uncertain climate Δ in Temperature Forecasts 0.7 Risk 0.5 0.4 Low Emissions % Δ in Precipitation Forecasts 0.3 0.2 Moderately High 0.07 Emissions 0.06 0.1 0.0 0 2 4 Δ in Temperature (°F) Conclusion <ICF International, 2009 in CCSP, 2009> 6 Probability Sensitivity Probability 0.6 0.05 0.04 Low Emissions 0.03 0.02 Moderately High Emissions 0.01 0.00 -15 -10 -5 0 5 10 15 20 Δ in Precipitation (in.) 118 118 Rationale Uncertain Survival for Low Emissions 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Uncertain Survival for Moderately High Emissions Expected Confidence Interval 1.0 0.8 0 20 40 60 Median Life (years) Conclusion 80 100 Probability Risk Probability Causes Sensitivity After 2,500 Simulations 0.6 0.4 Confidence Interval 0.2 Expected 0.0 0 20 40 60 80 100 Median Life (years) 119 119 Rationale Causes Median Life Current: 50 years Low Emissions: 50 yrs Mod. High Emissions: 49 yrs Sensitivity Conclusion Uncertain Median Life Probability Risk 90% CI [46,53] [45,54] 0.20 0.18 0.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 Low Emissions Moderately High Emissions 40 45 50 55 Median Life (years) 60 120120 Rationale Uncertain Life Uncertain EUAC Current: $2,328 Causes 90% CI [$2,286,$2,394] [$2,273,$2,394] Low: $2,328 Mod. High: $2,343 Sensitivity Conclusion Probability Risk Uncertain Median EUAC 0.20 0.18 0.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 $2,200 Low Emissions Moderately High Emissions $2,300 $2,400 EUAC ($) $2,500 121 121 Rationale Probability Future Average Life or EUAC < Current Average Life or EUAC Causes Low Emissions: 48.8% chance Sensitivity Mod. High Emissions: 51.6% chance Risk Conclusion 122122 Rationale Causes Sensitivity Risk Suppose assessing needs for 10 year planning horizon If assumed 50 year life then would not set aside funds for 30 year old bridge If consider risk of ‘failure’ then would expect to need P(‘Failure’ within planning horizon)*Cost [1-S(30+10)] * Replacement Cost = $16,712 for Low Emissions = $16,917 for Moderately High Emissions Conclusion 123 123 Rationale Risk of programming the wrong project Assume ranking based decision on ΔURSL Causes 𝑈 = 1.1659 ∗ 1 − 𝐸𝑋𝑃 −0.0195 ∗ 𝑅𝑆𝐿2 1 0.9 Sensitivity 0.8 Risk Utility 0.7 0.6 0.5 0.4 0.3 0.2 0.1 Conclusion 0 0 2 4 6 8 10 Remaining Service Life in Years 124124 Rationale Causes Sensitivity Risk Suppose ranking replacement projects for 10 year planning horizon based solely on RSL If assumed 50 year life then would estimate no utility for replacing 30 year old bridge Considering risk for either emission scenario… Expected ΔU P(Max Benefit) P(Benefit) +25 18.0% 31.7% Conclusion 125 125 Rationale Causes Sensitivity Risk Full Demonstration of Example Programmed into Spreadsheet Conclusion 126126 Rationale Causes Sensitivity Risk Conclusion ▪ Need to incorporate uncertainty ▪ Causes of uncertainty ▪ Methods for assessing uncertainty ▪ Sensitivity Analysis ▪ Risk Analysis ▪ Recommended to move towards probabilistic planning and management framework 127 127 128 129 To how many of these can you answer “yes”? LONG TERM VIEW Does the agency now feel confident in publishing life expectancy estimates, and using them to evaluate and anchor budgetary requests? Do senior managers have confidence that they know how much it will cost in the long term to sustain the desired level of service? Do outside stakeholders agree with management estimates of the long-term cost of sustaining the desired level of service? Do senior managers and stakeholders know what level of service can be sustained under current or proposed future funding levels? 130 To how many of these can you answer “yes”? TRANSPARENCY Is there a public comparison of forecast vs actual life expectancies? Are actions taken in response to life expectancy estimates and findings, and do stakeholders know what these actions are? Are comparisons routinely and publicly made of the agency’s performance against peer agencies, and against itself over time? 131 To how many of these can you answer “yes”? LEVELS OF SERVICE Can the agency accurately measure, track, and publish the level of service it is currently providing? Are life extension and replacement decisions accurately timed to avoid interruptions in service while minimizing costs? Is the agency reducing the annual number of traffic disruptions due to planned and unplanned maintenance, repair, and replacement activity? 132 To how many of these can you answer “yes”? EFFICIENCY Is the agency improving in its quantitative performance, in relation to the cost of providing the desired levels of service? Can the agency show, from its actual data, that its more refined timing of life extension and replacement actions is saving money, relative to earlier practice? Does the agency routinely compute, and effectively communicate, the life cycle costs of its services? Are these costs showing a clear trend of improvement? 133 To how many of these can you answer “yes”? AGENCY COMPETITIVENESS Is the agency using its asset management information as a competitive weapon to secure adequate funding? Are legislators confident that the agency is doing everything it can to control costs? Is the agency able to maintain adequate funding levels over time, in the face of competing uses of the money? 134 To how many of these can you answer “yes”? CONSTRUCTIVE RELATIONSHIPS Is the agency working actively with outside stakeholders on strategies to maintain and enhance the level of service provided to the public? Do outside stakeholders understand how their own interests are served by maintaining the agency’s level of service objectives? Do legislators and funding bodies rely on the agency’s models of the relationship between level of service and funding? 135 136