OR/17/009 General discussion

Tye, A M, Kirkwood, C, Dearden, R, Rawlins, B G, Lark, R M, Lawley, R L, Entwistle, D, and Mee, K. 2017. Environmental factors influencing pipe failures. British Geological Survey Internal Report, OR/17/009.

The value of the model outputs

The major aim of the project was to assess whether incorporating geological and environmental factors into models of pipe failure, water companies could develop greater understanding of their pipe networks. This may enable them to consider ways through which greater resilience can be built in, particularly with respect to a changing climate and increasing population. Typically water companies assess the current condition of their pipe assets by looking at age and increasingly internal camera assessments. Our approach is complementary and looks spatially at the distribution of pipe failure with respect to the density of pipe, and links these to geological, topographical and environmental factors. Thus the model produces spatial information of where these factors may have the greatest impacts on the pipe network. This is achieved through:

1. Interpretation of lurking variable plots allows an assessment of areas of the YW region where the model underperforms, allowing exploration of possible other factors that are causing damage to the network. Where under-prediction coincides on both the X and Y axes, reasons can be more easily identified. For example, in this work the models for different pipe materials consistently under-predicted pipe failure in areas associated with the coal measures and for cast iron, one area associated with lacustrine deposits. There are valid reasons (subsidence, poor load bearing strength) why these areas may have been identified.
2. The identification of significant model co-variables allows us to understand those factors that are having an effect on the network. Whereas the continuous variables produce a ± coefficient, greater interpretation is required for the categorical variables as it was not always a linear response.
3. A combined heat map can be produced by combining all the coefficients for each 100 x 100 m cell to show where the pipe network is at greatest risk.
4. The coefficients from the significant model co-variables from the final sequential model can be used in the production of individual heat maps which can help explain the factors contributing to the combined heat map. Thus by combining the coefficient with the categorical class or the continuous variable number for each 100 x 100 m cell, the areas in which individual covariates could impact the pipe network can be assessed spatially across the YW region. In addition, by taking the highest and lowest coefficient from all the covariates and standardising the colour scheme we can also compare the impacts of the covariates on the pipe network.

What we have learnt

Using the approaches outlined above, the following are the key points from analyses of the 4 pipe networks for YW:

1. For the YW region, non-geological factors generally had the greatest impact on pipe network failure including factors associated with road networks, water source and the number of dwellings.
2. The co-variables identified in the expert elicitation were usually found to be significant at P<0.05, demonstrating that YW had good knowledge regarding reasons for failure within their pipe network. The inclusion of factors identified through the Expert Elicitation exercise always improved the Null model. However the inclusion of further environmental and geohazard factors (e.g. dwellings, Sulphide/sulphate) resulted in improvements to the Expert Elicitation models.
3. Where geological and topographic factors were important these included slope for the concrete and clay networks which with gravity and weight obviously produces stress on the network. Solubility was important for the concrete pipe, suggesting subsidence in the Rippon area was a major source of failure. Sulphate and sulphide was important for cast iron, identifying partly the coal measures.
4. Some geological units appeared to cause problems for the pipe network beyond those accounted for in the list of co-variables. In particular these included the coal measures where subsidence may occur and one area of the lacustrine clays associated with the Glacial Lake Humber. Lacustrine clays typically have poor loading capacity.
5. The continuous variables were relatively easy to interpret as to their role in pipe network failure, whereas the use of the geohazard categorical variables did not always provide linear responses.
6. Shrink swell and compressible ground are the two geohazards often cited as having major impacts on pipe network failure and both were found to be significant (P<0.05), but the categorical coefficients obtained were non-linear. In addition, both datasets were not tested sufficiently. The YW region did not have a Class 5 region (Highly plastic soils) for shrink swell clays so no estimation of the most extreme shrink-swell clay soils could be made. For compressible ground conditions where pipe networks did pass through the highest class of risk, no known failures were found, which resulted in very low coefficient values. The interpretation of the categorical variables was therefore difficult. However, we suggest that the non-linearity of the coefficients obtained for the classes within these datasets may indicate broad ranges of different soil types and their specific properties which determine the settlement and deflection of different pipe materials in the soil. The amount of clay in the soil and it’s type are fundamental to shrink swell and compressibility but are also fundamental to processes that enable support to the pipeline. This needs to be examined further.
7. Some of the geohazard datasets needed interpretation because of the way they were created (e.g solubility, sulphide/sulphate). For example the solubility dataset could be split up into soft rocks that may dissolve (e.g. chalk and limestone) and those that may have soluble horizons causing subsidence (e.g gypsum containing rocks). For the sulphate and sulphide they could be split again. Individual datasets could be more appropriate and easier to use in some circumstances.

Review of work with yorkshire water, scottish water and welsh water

Presentations of results were made to Yorkshire Water during the course of these two grants. The meetings are reviewed here.

Meeting with Yorkshire Water — 22 June 2015

A meeting was held with representatives of YW on 22 June 2015 to gain feedback from the initial model results. Discussions focused on the possible causes of pipe failures caused by factors that were not included within the model and this particularly applied to the under prediction of the model in the SW of the YW region (Leeds-Bradford). These included:

• Surge demand
• Water pressure changes
• Water temperature & temperature change
• Source of water
• Drainage
• Climate

The area of model under prediction (Leeds-Bradford) is the largest urban area and this is likely to be where surge demand will most regularly occur. In addition, it is also a hilly region within the YW region and this may also cause greater changes in water pressure within the pipe network to occur. Both these factors are recognised as causing the blow out of pinhole corrosion to occur in cast iron pipes. Future data used in the model could include calculated ‘change of slope’ within a 100 x 100 m cell as well as mean slope to consider these pressure changes. A further possibility for the Leeds-Bradford area is that the source of water is different to the rest of the YW region (DWI, 2014). It was suggested that Leeds-Bradford may be served by reservoirs whilst much of the rest of the region by groundwater, and YW confirmed. These different water sources will have different temperatures and hydro-chemical variations. Groundwater temperatures should be constant, whereas reservoir water will vary depending on the season and weather. In addition the different chemistries (e.g. pH, SO₄^2-, Cl^-) of the water may have an effect on the internal corrosion of pipes. Data on temperature and pressure is held by YW for the Distribution Management Areas (DMA). There are 2300 DMA’s each serving between 800 and 900 properties and these could be included within the model.

The major covariate in the models for pipe failure was linked to ‘C-roads’. YW cited the following as possible contributory factors. The major weaknesses within a pipe network are the join between lengths of pipes. The pipe network associated with C roads is usually dominated by a greater frequency of connections between the ‘water main’ and the domestic pipe. In addition, smaller diameter and thinner pipes may be used in much of this part of the network. Within the C-roads we suggested that poor drainage in the sub grade may encourage anaerobic conditions associated with ponding of water can lead to corrosion. YW state that the infill of trenches is generally limestone gravel from about 1970 onwards. However, no comments were made about drainage. This suggests that in future modelling different drainage factors such as change of slope angle and drainage x geology may be appropriate. Different data and information sets were discussed. BGS could use derived data from NextMap to calculate the change of slope and the SUDS dataset for drainage get away.

Climate features were also considered. Cast iron and ductile pipes have peak bursts during the winter — December to February relating to low temperatures. Plastic pipes tend to fail more often during the summer. If the failure data is dated then an assessment of climate on bursts could be done within the model. Soil Moisture Deficit (SMD) was also mentioned and this should be feasible.

The causes of failure associated with the plastic pipe network appear to be related to vibration, slope and clay. All three road types were identified, with C roads > B roads > A roads in the ranking, suggesting that the larger better constructed roads have a lower effect. Slope may play an effect through gravity distorting the pipes. Interestingly, clay was identified as a proxy of A- resistivity. The hardness of and rigidity of dry clay or its contributions towards ground movements may be significant. Hardness may cause chaffing of the pipeline with vibrations. Interestingly there was a negative correlation with compressible ground suggesting that a pipe in slightly giving material may be slightly protected. This may also be why there was a negative correlation between plastic pipe failure and shrink-swell clays.

Meeting with Yorkshire Water — 14 July 2016

A further meeting was held with YW where improvements in the cast iron and plastic models created in Grant NE/NO13026/1 were presented as well as results from the waste water clay and concrete models. Discussions were had regarding explanations of results. Main points regarding results of pipe networks were:

1. C-roads more likely to be in road whilst A- and B-roads are in the pavement if possible. In addition because of the type of road there will be differences in size and number of connections. In C-roads it’s likely to be a ‘distribution’ network whilst larger road we have the principal mains.
2. Unlikely for pipes put in shrink swell soils to be differently engineered — Current YW models do not have this differentiation in their models built a basic soil type factor
3. Extra protection would be given to pipes if peat is present
4. Pipe depth is a factor that hasn’t been included within the model because of access to suitable data. Standard depth for cost and temperature. More variability for concrete waste water. There is scope for shallower depth with plastic pipes. It will help to reduce carbon footprint (digging) being able to reduce depth to which pipes are laid. Waste pipes often installed using micro-tunnelling technologies these days rather than open cut.

Meeting with Scottish Water — 27 June 2016

A lot of decisions are currently made according to beliefs rather than evidence. There is a drive in Scottish Water to become more data driven/evidence based. They are aware of a correlation between pressure and pipe failure — e.g. 10% reduction in pressure results in 14% reduction in failure rate (but, it may just prolong inevitable failure by corrosion). In their efforts to reduce pipe failure, 50% of the reduction they have been able to achieve has been as a result of pressure management. The rest is mains rehab and operations management. They do not currently engineer to account for geological conditions such as shrink swell. Pipe systems are ‘off the shelf’, not specific to ground conditions.

Expert Elicitation

An expert elicitation exercise was undertaken with Scottish Water so that discussion of their results would not bias their opinions on what they considered were the principal reasons for failure within their pipelines. Compared to YW, Scottish Water there was more slightly more focus on weather effects (Scotland generally having longer and colder winters). These are the results of the Expert Elicitation process.

Cast iron:

1. Age (exacerbates all other factors — an interaction effect?)
2. Pressure (pressure transience rather than constant pressure)
3. Ground temperature
4. Weather (seasonality, cold and wet winter vs drying out of ground in summer)
5. Ground heave (failure by ring splits, ring fractures)
6. Water source (ground, surface, chemical treatment — internal corrosion > external corrosion)
7. Road vibration (construction sites, building and piling — in theory 600 mm of cover makes this negligible but they are suspicious)
8. Contaminated ground (e.g. High rate of corrosion/pitting at Innerleithen due to copper contamination?).

Plastic:

1. Installation error
2. Pressure transience
3. Presence of hydrocarbons possibly (e.g. peat)

Asbestos cement:

1. External factors (e.g. pH, water chemistry, soil types)
2. Mechanical joint failure — installation problems or corrosion of nuts and bolts (see cast iron)

Pre-stressed concrete:

1. Catastrophic join failure

Clay pipes:

1. Root infiltration
2. Ground distortion/disturbance (rubber seals will pop out)

General additional hazards:

1. Ground water infiltration may be an issue — they mentioned mining areas and red ochre — acid mine drainage?
2. Peat — pipe buoyancy and mobility, and hydrocarbons
3. Running sands an issue for sewers
4. Mine collapse an issue but rare

Suggestions from Scottish Water during BGS presentation:

1. They were interested if we had included a flooding layer as a predictor variable in our models, with reference being to ground water infiltration of sewer system.
2. Saline infiltration was an issue on the East Coast, where rising mains are metallic.
3. They were interested in our use of number of dwellings per cell and mentioned work that they had done looking at social demographic/class and sewer blockages.
4. C road influence may be due to construction activity and third party damage. Possible pipe failures as a result of contrasting ground conditions between made ground under the road and natural ground beyond.
5. We could perhaps use CEH’s land use classification as a predictor (but land use may effectively already be explained in what we have used).
6. With regard to dwellings — is it possible that our burst data includes bursts in minor house-feeding pipes, which have been mistakenly appended to the mains?
7. It was suggested that the model could be validated by splitting the data into blocks of different age and comparing the resultant models.

Meeting with Welsh Water — 22 July 2016

An expert elicitation process was carried out with the main comments for cast iron being similar to those from Scottish Water and Yorkshire Water. Again weather, particularly the autumn period when a greater number of failures are reported was mentioned.

Cast iron
For the cast iron network the results are shown below

1. Climate — winter freezing
2. Soil moisture deficit
3. Corrosive soils
4. Age
5. Pressure transience
6. Joints