5 5.3. Methodology 115 5.3 Methodology Our goal is to describe the deterioration of sewer mains based on a historic set of inspection data. To achieve this we calibrate DTMCs that quantify the probability of a pipe (from the historic dataset) being in a condition class given the age of the pipe. These DTMCs represent cohorts of pipes, i.e., they are trained with data from pipes that share similar characteristics. An overview of the four steps we follow includes: Step 1, pre-processing the data (cleaning); Step 2, defining cohorts; Step 3, creating a discretised table per cohort, which serves as the input data to calibrate the DTMC; and Step 4, calibrating the DTMC. The details about each of the steps are provided in the following sections. 5.3.1 Step 1: Data pre-processing Our work is based on the case study detailed in Section II.4.3. The dataset for each sewer main includes information on inspections, such as (i) unique inspection identifiers, (ii) inspection dates, (iii) damage sizes, (iv) damage codes, (v) damage classes (described by ϑ), and (vi) the damage’s relative position. Each damage code is associated with a damage class. For this chapter, we exclude data on pipes built before 1920 and those with missing or erroneous construction year information. 5.3.2 Step 2: Definition of cohorts To account for explanatory variables beyond pipe age in deterioration modelling, it is necessary to construct cohorts (groups of sewer mains with similar characteristics) and calibrate a DTMC for each cohort. Table 5.1 presents six cohorts of interest and the number of pipes with specific characteristics, expressed as a fraction of the total inspected pipes. However, a drawback of defining cohorts is the potential for small subsets (e.g., cohort PR), which may not be statistically representative. Table 5.1: Cohorts of interest, fraction (%) of total inspected pipes (25,507). Cohort name Description Fraction (%) CMW Material: concrete & Content: mixed and waste 59.29 CR Material: concrete & Content: stormwater 12.85 PMW Material: PVC & Content: mixed and waste 18.88 PR Material: PVC & Content: stormwater 7.89 CdL Material: concrete & width <500 mm 50.16 CdG Material: concrete & width ↔500 mm 22.02 5.3.3 Step 3: Discretised table We construct a discretised table for each cohort (Table 5.2) and damage code to serve as input for calibrating the DTMCs. The state of a sewer pipe is identified
RkJQdWJsaXNoZXIy MjY0ODMw