Pipe Condition Assessment and Burst Prediction: From Pit Depth to the Economic Renewal Date

How the main materials fail

The UK network is dominated by five materials, each with a characteristic failure mode:

In practical terms, "age" is a proxy. A 1900 cast iron main in benign clay with a thick wall may outlast a 1970 asbestos cement main in sulphate soil. Condition assessment replaces age with the variables that actually govern: wall loss, residual strength, soil aggressivity, loading and operating pressure.

Corrosion pit growth and residual strength

External corrosion of iron pipe proceeds by pitting whose maximum depth grows with time as a power law:

With k = 1.0 mm/yrⁿ and n = 0.5, a pipe reaches 10 mm of pit depth after 100 years; in an aggressive soil with k = 2.0 the same depth is reached in 25 years. For a grey iron main with an original wall of 12 mm, 10 mm of pitting leaves 2 mm of sound metal at the worst point. Residual pressure capacity at that point follows from the hoop stress relation with the remaining wall:

At 6 bar (0.6 MPa) in a 150 mm pipe with 2 mm remaining: σ = 0.6 × 150 / (2 × 2) = 45 MPa, which is well within grey iron's tensile strength of 150 to 200 MPa. That is why pure hoop stress rarely explains cast iron failures. The actual mechanism is fracture: a corrosion pit is a stress concentrator, grey iron has low fracture toughness, and the loading that breaks it is bending from ground movement, traffic and thermal contraction, superimposed on the pressure stress. Combined stress at a pit tip exceeds the fracture criterion long before the hoop stress alone would. That is also why small-diameter cast iron fails circumferentially in cold, dry weather and why burst rates spike when soil moisture changes.

Burst-rate models

For planning, the individual-pipe physics is aggregated into statistical models of burst frequency for a cohort of pipes with shared material, age, diameter and soil. Shamir and Howard's classic form is exponential growth in failures per unit length:

Modern utilities fit richer models, in particular survival and hazard models (Weibull or Cox proportional hazards) that treat each pipe's time to next failure as a random variable conditioned on covariates: material, diameter, age, previous bursts, soil class, pressure, traffic loading. Two empirical regularities matter for interpretation. First, previous bursts are the strongest predictor of the next one: a pipe that has failed twice in five years is in a different hazard class from its age cohort. Second, the hazard is not monotonic for all materials; early-life (installation) failures, a long stable middle and a rising tail give a bathtub shape, so a young PE cohort with a burst cluster is signalling a construction problem, not age.

Outputs are a ranked list of pipes by predicted failure rate and, summed across the network, a forecast of total bursts under different renewal rates, which is the number that regulatory performance commitments on mains repairs and supply interruptions are built on.

Condition survey methods

Models trained on burst records predict cohorts; condition surveys calibrate them and decide individual pipes. Methods, roughly from cheapest to most informative:

• Soil surveys: resistivity, pH, redox, sulphide and moisture along the route classify corrosivity and set k in the pit-growth law; they cost little and sharpen the cohort model considerably.
• Exhumed sample analysis: a pipe cut-out from a repair measured for pit depth distribution and graphitisation; a few samples per cohort anchor the model to reality.
• Acoustic wall-thickness assessment: the speed of a pressure wave between two sensors depends on the pipe's effective stiffness (the same relation that governs leak correlation), so a reduced wave speed indicates lost wall; averaged over the section but non-intrusive and fast.
• Inline inspection: tethered or free-swimming tools carrying magnetic flux leakage, remote field eddy current or ultrasonic sensors report wall loss metre by metre; the tool of choice for trunk mains where a failure is unacceptable and the pipe can be accessed.
• Pressure transient logging: identifies the surges and pressure cycling that drive fatigue failures, which no wall-thickness method sees.

Distribution mains are too numerous for inline inspection, so the usual strategy is cohort modelling calibrated with soil data and samples, with intrusive inspection reserved for trunk mains and critical crossings.

The economic renewal date

The decision to replace is economic. Let the burst rate grow exponentially at A, each burst cost c_b (repair, water lost, interruption and any regulatory penalty), replacement cost C_r, and the continuous discount rate be ρ. Replacing at time T has present value of burst costs up to T plus the discounted replacement; differentiating and setting to zero gives the optimum:

The rule says: replace when the annual cost of bursts (c_b N) has grown to equal the annual cost of money tied up in the replacement (ρ C_r). It ignores the residual burst rate of the new pipe and later renewal cycles, which shorten T* slightly, but it captures the first-order behaviour and is easy to run across every cohort.

Worked example. A 1 km length of 100 mm grey iron main currently suffers 1.0 bursts per year (N₀ = 1.0), with each burst costing £5,000 all-in. Repair records give A = 0.05 per year (doubling every 14 years). Replacement by pipe bursting costs £300 per metre, so C_r = £300,000, and the company's real discount rate is 3.5 percent.

• ρ C_r = 0.035 × 300,000 = £10,500/yr; c_b N₀ = 5,000 × 1.0 = £5,000/yr
• T* = (1/0.05) × ln(10,500 / 5,000) = 20 × ln(2.1) = 20 × 0.742 = 14.8 years

Despite a burst a year, the main is not yet economic to replace. If the burst cost rises to £10,000 (a main under a trunk road, or with a high supply-interruption penalty), the bracket is 1.05 and T* is 1 year: replace in the next programme. If the burst rate is 2 per year at £5,000, the bracket is 1.05 again. This sensitivity shows why the ODI rates set in the price review change renewal programmes so directly: they change c_b. It also shows why a main with repeat bursts under a road is replaced long before a main with the same burst rate in a field.

From assessment to programme

A renewal programme assembled from the economic rule and the hazard ranking has three further layers. Consequence: a trunk main feeding a hospital or a single-feed town is renewed on risk, not on burst cost alone. Opportunity: a highway reconstruction or a sewer scheme in the same street cuts replacement cost and advances the date. Method: once a main is selected, the trenchless rehabilitation guide covers choosing between bursting, lining and open cut, which in turn feeds back into C_r.

Output should be a ranked, costed list with a forecast burst rate under each funding level, because that is what the regulator will ask for and what the operations team needs to plan. Where renewal interacts with pumping, pressure zones or treatment (a smoother, larger main changes the hydraulics downstream), process and hydraulic design support closes the loop before construction.

How to assess a cohort of mains for renewal

1. Build the cohort. Group pipes by material, diameter band, installation era and soil class from asset records and a soil survey.
2. Fit the burst model. From repair records, fit burst rate against time for each cohort (exponential or survival model) and identify pipes with repeat failures as a separate high-hazard class.
3. Calibrate with samples. Measure pit depth and wall loss on exhumed cut-outs; adjust the model where it disagrees with physical evidence.
4. Cost a burst. Repair, water lost, customer interruption, traffic management, regulatory penalty; vary by location so that consequence is in the number.
5. Compute the economic date. Apply the renewal rule with the cohort growth rate, current burst rate, replacement cost and discount rate; list pipes by years to T*.
6. Overlay risk and opportunity. Advance critical mains and those with co-funded street works; defer low-consequence mains with spare time; publish the ranked programme with its forecast burst rate.