Appendix 'C' - Expert Encoding Package

RE: Bayesian Subgrade Modeling Application.

Dear Mr. MacLeod:

Materials Section, Technical Policies and Standards Branch is undertaking a subgrade performance modeling exercise in order to learn the Bayesian Application and Software and to develop a first generation performance model for unpaved roads.

The Bayesian Methodology and software was developed for the C­SHRP LTPP project in order to develop performance models for pavement overlays. The three models developed were ride, rut depth and fatigue. The power of the Bayesian methodology is that it allows models to be developed early in an experiment with very little data. As more data is obtained the model can be adjusted to reflect the new information.

The basic process consists of defining the model that is to be developed in terms of a dependent variable (What you want to predict or measure.) and any number of independent variables. The judgment of experts and initial data is used to develop a first phase model which allows you to look at the significance of each variable as well as give early predictions. Future data collection can be planned to improve the model with time.

Your assistance is being requested to provide the expert judgment related to subgrade performance in Saskatchewan. The attached "Expert Encoding Package" provides all of the required information and the forms that you are requested to fill out.

It is anticipated that it will take you I to 2 hours to complete the task.

We would appreciate a response, either by return mail or fax, by May 30, 1995. This would consist of returning all 4 Tables.

Thank you for your assistance in this project. I will provide you a copy of the first phase analysis when it is complete. If you have any questions, please give myself or Randy Schmidt, who is leading the analysis, a call.

Allan Widger

copy: Randy Schmidt

EXPERT ENCODING PACKAGE INSTRUCTIONS

Unpaved Road Research and Modelling Project First Generation Bayesian Performance Model

BACKGROUND

Early in the spring on 1994 Materials Section, Saskatchewan Highways and Transportation (SHT) was given the task to investigate the issue of subgrade design and characterization as it relates to performance of unpaved roads. The intent was to take a "quick and dirty" look at the factors that impacted on subgrade performance and to attempt to collect the required field data needed to develop some kind of performance model. SHT wanted to develop a short and long term plan for the analysis of unpaved roads and a method of performance prediction under different loading conditions.

The study was initiated because concentrated grain haul from elevators to inland terminals was causing significant damage on many gravel and Thin Membrane Surfaces (TMS) and it was hoped that some method could be developed to predict the performance of these roads. It was hoped that in the future, hauling could be controlled to acceptable levels to reduce or eliminate the damage. With the changes in the Crow grain transportation subsidy and the move to abandon rail lines and rationalism the elevator systems the problem of unpaved road performance has taken on a very high profile significance in Saskatchewan.

The issue of subgrade performance had not been looked at by SHT since the 1960's and 1970 so it was hoped that new technology could be applied to subgrade design, analysis and performance modelling.

Part way through the data collection process SHT had the opportunity to take part in a TAC sponsored Bayesian Methods and BSTAT Software training Session. A proposal was put forward and accepted by TAC for a joint TAC ­SHT Bayesian Application Project to use the newly developed Bayesian software as a tool in developing a subgrade performance model.

THE PROBLEM:

The problem was defined as one of being able to characterize a subgrade and predict the type of loading that it could carry.

Unpaved roads, or roads with only surface treatments, dependent on the subgrade materials for performance. The factors that control the subgrade performance and the interaction between factors, have never totally been understood and there is no reliable method available to evaluate or predict the load carrying capability of a subgrade throughout the year.

The influence of truck loading can be incorporated in pavement design but its impact on unpaved is not understood. Truck loading factors that must be considered are the type of truck (number and configuration of axles), gross vehicle weights, the axle weights, the number of repetitions and the frequency of repetitions. A future consideration could be the tire inflation pressure.

There have been other developments related to materials characterization in lab and field testing equipment and instrumentation that now allow the measurement of insitu strength, soil matrix suction, moisture content and temperature over time and changing environmental conditions.

There is no clear definition of what is meant by subgrade performance and how subgrade failure can be defined. In order to develop a model it is necessary to define performance in terms of some dependent variable that can be measured in the field.

PROJECT OBJECTIVES

The project has a number of objectives

a. Clearly define all of the variables that characterize subgrade materials or that contribute to the performance of the subgrade for paved and unpaved roads. Prioritise the variables in an attempt to limit the number that have to be studied or identify their contribution to performance.

b. Capture the experience and knowledge gained from past and current studies and research in order to develop a plan to meet future needs.

c. Combine all the related variables into a reasonable number of components that control subgrade design and performance in order to develop a manageable model.

d. Develop short and long term data collection, handling and analysis techniques that are practical and economical and can be used to develop a new subgrade characterization, design and performance evaluation method.

e. Develop a first generation performance models on which to build.

f. To learn the Bayesian process and attempt to incorporate it into the research project design and data analysis.

DEFINING VARIABLES

The first step in designing a Bayesian Modelling Project is to ask a number of experts to determine what variables they feel contribute to the problem and to prioritise the variables.

A list of variables was developed for the project but in order to test the full utilisation of the Bayesian process you are requested to provide your input into variable identification. A list of variables is given in Table I of the attached Encoding Package. You are asked to add any missing variables to the list. You are then asked to rate the variables on a scale of I to 100 with I having the least contribution to subgrade performance and 100 being the most significant factors.

Please complete Table I before proceeding.

If your were to define subgrade performance or failure in terms of a dependent variable that can be predicted and measured what would you suggest. Table 2 gives a list of possible performance measures. Please add any additional factors that you feel could be defined or measured that could define performance or failure.

Again rate them on a scale of I to 100 with 100 being the best measure of performance.

Please complete Table 2 before proceeding to the rest of the package.

RESEARCH PROCEDURE

In a normal Bayesian project an experiment or data collection program would be designed to collect data on the top ranked variables and dependent variable and use this data to build the model.

The data collection for this project had been designed to collect data over a short period of time in order to get a feel for what data could be collected, how to handle and present the data and what data was meaningful. Two construction haul roads were selected where a large quantity of aggregate would be hauled out over a length of gravel road in a short time span. A decision was made to collect data that could be readily obtained at set location before and during the haul. Visual inspections were made and additional testing was completed at specific locations where changes in the road condition occurred.

Performance was being measured in terms of deflections and visual condition while the following data was collected; daily loading, nuclear density, insitu density, moisture content, unified classification, pocket penetration on spoon samples, and dynamic cone penetrations (DCP).

The research project data collection was well underway before the Bayesian project was started so it was necessary to select a dependent variable that had been measured and build the model based on what data had been collected.

Deflection was selected as the dependent variable to be modelled. Loading data was very good since we had the weight tickets from all the trucks and could define any loading variable desired. The subgrade data was compiled in order to determine how best to include it in the analysis. It was found that the basic material properties gave a shotgun scatter and that no trends could be defined over time. In order to define the subgrade in terms that an expert could relate to performance the data was re­assessed.

It was determined that the pop gave very consistent results that showed a number of trends. There was concern that experts would not be familiar with pop results so a decision was made to convert the pop results to CBR using the following equation developed by Kleyn and Van Heerden in 1975.

Log CBR = 2.628 ­ 1.273Log pop

The pop value is a measured of penetration in terms of mm/blow. Figure I shows the dimensions. The form of the equation is such that if the penetration per blow becomes very low (less than 3mm/blow) the CBR exceeds 100 and becomes meaningless so these values were recorded as a CBR of 100.

The data showed that in almost all cases there is a distinct change in strength from a strong surface layer to a weaker underlayer. The strong surface layer was defined as the crust . The crust could be assigned a thickness and strength value. The underlying material tended to have a uniform strength throughout the 1.2m depth measured. Figure 2 shows a typical pop result with the crust thickness and calculated CBRs. These relationships were developed for all tests taken during the study.

Definition of Variables

Our intent is to use your judgement combined with that of other experts in conjunction with the data collected on the two haul roads; therefore we have explicitly set the definitions and the same inference space across each variable to be comparable for both data sets.

The following discussion provides an overview of the variables on which we wish to encode your expert judgement. It is important that all encoders view the problem in the same way and make the same assumptions.

DEPENDANT VARIABLE

The dependent variable in the Bayesian model is rebound deflection in millimetres (recorded to 0.1 mm) as measured by a Benkelman beam under a 80KN standard axle. The road surface was generally very

hard so it was possible to brush away any loose material and to take a normal beam deflection. Measurements were taken in the outside wheel path but at least I m from the unsupported edge of the road. The readings were not corrected for temperature.

The deflections measured at the sites were distributed over a wide range. Your inference space should range from a low reading of 0.1 mm to a high reading of 9.5mm.

INDEPENDENT VARIABLES

The independent variables defined below all apply to a normal grid road standard with a 8m top width. Both sections are old roads that have carried gravel hauls in the past but normally would have ADTs of less than 100 vpd and very few trucks. The roads are level with a slight crown. No measurements were taken on curves and there were no hills. The grade height ranges from 1.2m to 1.8m with no standing water or groundwater discharge. There was no significant prolonged rainfall during the period of the haul and temperatures were in the normal summer range. Normal surface spot blading occurred during haul but there was no major blading or windrow of gravel.

The subgrade material varied slightly between holes and with depth. The material was a till with traffic gravel worked or pushed into the surface. The following properties would be typical of those measured over the sites.

You will be asked to estimate a deflection in between 0.1 and 9.5mm for all combinations of variables. Your estimate will be encoded on the encoding matrix in Table 4.

Crust Thickness

Crust thickness was defined as the depth or thickness at which the slope of the pop penetration plot took a sharp break indicating a drop in strength. A thickness of less than 50mm was considered to be no strength and only the strength of the subgrade would apply. The thickness range of 50 to 150mm will be encoded as 100mm and the thickness range greater than 150mm will be encoded as 175mm. Crust Strength

The crust strength is reported in terms of CBR with values in the range CBR40 to CBR70 being encoded as CBR55, the range CBR70 to CBR100 being encoded as CBR85 and above CBR100 encoded as CBR100.

Subgrade Strength

The subgrade strength is reported in terms of CBR with values in the range less than CBR15 being encoded as CBR7, values in the range CBR15 to CBR30 being encoded as CBR22, and values greater than CBR30 being encoded as CBR30. Total Loading

Total loading was recorded in terms of total Gross Tonnes which were hauled over the section throughout the term of monitoring. The trucks hauled at secondary weights and consisted of an almost even split of 5 axle semi­trailer loaded to 34.5t (steering 5500t, drive tandem 14.5t and rear tandem 14.5 tonne) and 6 axle semi­trailers loaded to 40.0t(steering 5500t. drive tandem 14.5t and rear tridem 20t) both having an EASL equivalent of approximately 4 EASLs per load.

The total loading was broken into three ranges. Less than 40,000t was encoded as 20,000t (approximately 540 loads, 3000 axle passes or 2000 ESALs), 40,000t to 120,000t was encoded as 80,000t (approximately 2000 loads, 12,000 axles or 8000 ESALs). and greater than 120,000t encoded as 150,000t (approximately 4000 loads, 24,000 axles or 16,000 ESALs).

Repetitive Loading The repetitive loading was recorded in terms of axles/day. Less than 220 axles/day was encoded as 120 axles/day

(20 loads, 800t or 80 ESALs), the range of 220 to 450 axles/day was encoded as 320 axles/day (60 loads, 2400 t,, or 240 ESALs) and greater than 450 axles/day was encoded as 500 axles/day (90 loads, 3600 tonnes, or 360 ESALs).

RELATIONSHIP BETWEEN DEFLECTION AND OTHER VARIABLES

In order that we might understand your reasoning in completing the encoding we ask that you express your expert opinion on the relationship of each variable to the predicted deflection. Table 3 shows a series of five graphs. We would request that you sketch the form of the relationship that exists between the deflection and the selected

variable. We are only looking at the trend than you feel occurs and are not interested in the scale. Please provide a short description of each relationship and an explanation of your reasoning. This information will be used if there is

a major divergence between experts. Comparing the trend plots and reasoning will help to show if all experts had the same understanding of the variables.

ENCODING PROCESS

You are now ready to proceed to Table 4 which is the encoding form for your expert judgement. Each cell in the matrix corresponds to a unique combination of the variables. We ask that you place your estimated deflection (between 0. 1mm and 9.5mm) in each cell. You may use whatever process you feel comfortable with in moving around and filling in the matrix. (It is suggested that you try to estimate each cell for its own unique conditions so that you don't get trapped in a pegging bias which results from trying to keep an equal relationship between all variables.)

PERFORMANCE CRITERIA

Your expert judgement, the expert judgement of your peers and the data collected at the two research sites will be used to create a first generation model of the following form.

As the last task we would ask that you provide your estimate of how failure could be defined in terms of a limiting deflection. You may select a single value for all conditions or different values for different variable conditions. These values may be shown on the encoding form in the place provided.

If you have any questions or need further clarification please call Allan Widger at 787­4858 or Randy Schmidt at 787­4835.

• Table 1
• Table 2
• Table 3
• Table 4

Return to Main Page