1. Helsinki University of Technology Systems Analysis Laboratory INFORMS Seattle 2007 Integrated Multi-Criteria Budgeting for Maintenance and Rehabilitation Policies at the Finnish Road Administration Pekka Mild and Ahti Salo Systems Analysis Laboratory Helsinki University of Technology (TKK) P.O. Box 1100, 02015 TKK, Finland

2. Helsinki University of Technology Systems Analysis Laboratory 2INFORMS Seattle 2007, Pekka Mild and Ahti Salo Road asset management in Finland n Finnish Road Administration (Finnra) –Central administration and 9 road districts –Maintenance, repair and investments mgmt –Research and development n Road network –78000 km of public roads –14000 bridges n Estimated asset value 21 billion USD –Around 4000 USD per capita –Annual funding around 850 million USD Road asset management research program 2003-2007

3. Helsinki University of Technology Systems Analysis Laboratory 3INFORMS Seattle 2007, Pekka Mild and Ahti Salo Programmed rehabilitation and reconstruction projects PavementsBridgesGravel roadsRoad equipment Day-to-day maintenance operations Winter-time operationsRoad surroundingsGravel roads How to allocate funds among road keeping products? Finnra must address multiple objectives in its policies Shift from technical maintenance to customer and service orientation New unified quality classes map levels of service All products impact the same road system No integrated management system to-date → static funding patterns Yet, sustainable development calls for dynamic (re)allocations Build an integrated framework for resource allocation Multi-criteria framework as the ”common language” among products Bring managers together to address future funding needs

4. Helsinki University of Technology Systems Analysis Laboratory 4INFORMS Seattle 2007, Pekka Mild and Ahti Salo Road district’s annual rehabilitation and maintenance budget Programmed rehabilitation and reconstruction projects Day-to-day road maintenance operations PavementsBridges Gravel roads Road equipment Winter- time Road sur- roundings Gravel- roads... 5 quality classes for all twig-level products 1 2345 High traf.Low... 1-3 sub-categories per product type => altogether 13 twig-level products compete for funding Road products and evaluation criteria ROAD SAFETY ENVIRONMENTAL IMPACT CUSTOMER SATISFACTION ASSET VALUE PRESERVATION

5. Helsinki University of Technology Systems Analysis Laboratory 5INFORMS Seattle 2007, Pekka Mild and Ahti Salo Value-focused evaluation of products n TKK-facilitated one-day workshop –10 experts from Finnra and Pöyry Infra Ltd. n Score elicitation –Intermediate scores by adjusting the shape of the value functions for each product –Maximum scores by comparing inter-product swings from the worst quality class to the best –These two phases repeated for all four criteria n Weight elicitation –Incomplete rank information about maximum swings under each criterion 1 2 3 4 5 class (j) 100 0 50 Customer satisfaction bridges 1 2 3 4 5 class (j) 100 0 50 Customer satisfaction bridges winter mnt. gravel rd.

6. Helsinki University of Technology Systems Analysis Laboratory 6INFORMS Seattle 2007, Pekka Mild and Ahti Salo Aggregate multicriteria value of products 1 2 3 4 5 quantity (q j ) class (j) bridges: quality class distribution Score times quantity 1 2 3 4 5 Road safety Environmental impact Asset value 1 2 3 4 5 Customer satisfaction bridges 2007 Year Weighted sum of scores

7. Helsinki University of Technology Systems Analysis Laboratory 7INFORMS Seattle 2007, Pekka Mild and Ahti Salo Deterioration and repair dynamics of products 1 2 3 4 5 quantity (q j ) class (j) 2007 Yr. 1 2 3 4 5 t + 1 $ 2007 Yr. t + 1 2008 1 2 3 4 5 t + n 2007 Yr. 200820092010201120122037 … … … n Products deteriorate towards worse quality classes over time n Repairs raise quality

8. Helsinki University of Technology Systems Analysis Laboratory 8INFORMS Seattle 2007, Pekka Mild and Ahti Salo Optimal resource allocations n Maximize the long-term sum of all products’ multicriteria value –Time horizon of 30 years with 3% p.a. discount rate –Budget constraints and quality targets –Decision variables: repair actions and levels of maintenance operations »Number of quality class 1 bridges repaired to class 4 in year 2008 »Kilometers held at winter maintenance quality class 3 in year 2012 –Repair and deterioration dynamics captured by linear constraints n Different weights suggest different optimal allocations –Sample the feasible weight set determined by the rank-ordering

9. Helsinki University of Technology Systems Analysis Laboratory 9INFORMS Seattle 2007, Pekka Mild and Ahti Salo Key results for management n Which resource allocation policies maximize the long-term multicriteria value of the whole road system? –Which products call for more funding when customer satisfaction becomes a key priority? –What do criteria weightings imply for the products’ funding needs? –What is the expected interim/terminal quality distribution of the system? n What is the ”pecking order” of the products? –Which products gain/lose funding when the overall budget is changed? –Which products gain/lose funding first and which later? –What do different weightings imply for the ”pecking order”?

10. Helsinki University of Technology Systems Analysis Laboratory 10INFORMS Seattle 2007, Pekka Mild and Ahti Salo Integrated platform for collaborative management of the entire system $ … …

11. Helsinki University of Technology Systems Analysis Laboratory 11INFORMS Seattle 2007, Pekka Mild and Ahti Salo Client feedback n Best project award in Finnra’s road asset management research program n ”An innovative tool for thinking and communication” –Antti Rinta-Porkkunen, Director of the South-East Finland road district n ”Framework to bring the managers of separated products to facilitated interaction and give them fresh insights about the aggregate system” –Vesa Männistö, Senior Consultant, Pöyry Infra Ltd. n Enthusiasm for optimization and decision analysis at Finnra

12. Helsinki University of Technology Systems Analysis Laboratory 12INFORMS Seattle 2007, Pekka Mild and Ahti Salo Novel methodological elements in our case  From technical condition-focus to value-focus –Explicit value models for quality classes From product orientation to portfolio optimization –Incomplete preference information through rank-orderings ŽFrom static budgeting to long-term allocations –Integrated repair and deterioration dynamics of products From turf-fights to collaborative learning –Interactive work-shop with ’on-the-fly’ computations

13. Helsinki University of Technology Systems Analysis Laboratory 13INFORMS Seattle 2007, Pekka Mild and Ahti Salo Towards integrated sustainable planning n Infrastructure & transportation asset management –Consumes enormous financial resources globally –Has far-reaching impacts on societies, industries and individuals –Involves multiple objectives, long planning horizons, high uncertainties n There is major untapped potential for Decision Analysis –Value-focused analysis of individual products and product portfolios –Explicit recognition of stakeholders’ interests and preferences –Use of DA models as vehicles for enhanced communication –A paradigm shift towards integrated collaborative planning

14. Helsinki University of Technology Systems Analysis Laboratory 14INFORMS Seattle 2007, Pekka Mild and Ahti Salo Thank you! Questions?

15. Helsinki University of Technology Systems Analysis Laboratory 15INFORMS Seattle 2007, Pekka Mild and Ahti Salo Appendix: LP-model formulation (1/3), variables & dynamics n Decision variables (product i, class j, year t) –Quantity distribution: –Amount (kilometers, units) moved from j to j’: n Linear repair and deterioration dynamics –Percentage of quantity deteriorates, i.e., drops to in one year – for all maintenance operations products –Linear constraints –Slightly different constraints for boundary states (1 and 5) –Set of allowed state transitions can be restricted product-wise Remains in class jDeteriorates from class j+1 Moved upwards from j to j’ Moves that arrive at j from below j’

16. Helsinki University of Technology Systems Analysis Laboratory 16INFORMS Seattle 2007, Pekka Mild and Ahti Salo Appendix: LP-model formulation (2/3), objective function n Evaluation score (product i, class j, criterion k) n Value of distribution (product i, criterion k, year t) –q ij (t): quantity of product i in class j in year t n Overall value of distribution (product i, year t) –w k : weight of criterion k (incomplete weighting w  S w ) n Overall value of all products (year t) –Sum of all products’ distributions’ overall values n Total overall value discounted over 30 years –Objective function in the optimization

17. Helsinki University of Technology Systems Analysis Laboratory 17INFORMS Seattle 2007, Pekka Mild and Ahti Salo Appendix: LP-model formulation (3/3), costs & constraints n Costs –Programmed repairs (i  REP): unit cost per move is –Maintenance operations (i  MNT): unit cost of service level is – for i  MNT (shifts are free but the resulting quantity comes to cost) n Budget constraints –Budget constraints can be set also for any subsets of products or moves n Examples of other constraints –Gradual change –(Dynamic) target thresholds for distributions –E.g., share of poor-conditioned (class 1) bridges must be below 1% in year 2015