A wall-flow filter is a ceramic honeycomb with alternate channels plugged at opposite ends, so that exhaust must pass through the porous walls. Soot is trapped first within the wall and then as a cake on its surface, and collection efficiency by mass is very high once the cake has formed (Konstandopoulos and Johnson, 1989). The pressure drop rises as soot accumulates (Konstandopoulos et al., 2000), so the soot has to be burned off. With oxygen alone this requires temperatures far above those in the exhaust of a bus in city traffic.

Two families of solution exist. Passive systems lower the temperature needed. The best known places an oxidation catalyst ahead of the filter to convert nitric oxide to nitrogen dioxide, which oxidises soot at much lower temperatures than oxygen does (Cooper and Thoss, 1989; Allansson et al., 2002); others coat the filter itself with catalyst (van Setten et al., 2001). Active systems raise the temperature when needed, by a fuel burner, electric heating or fuel dosed over a catalyst. Passive systems are simpler and cheaper but depend on the exhaust being hot enough for long enough, and on a sufficient ratio of nitrogen oxides to soot. A general account is given by Majewski and Khair (2006).

Regeneration removes carbon. It does not remove the metallic residue of lubricating oil additives, chiefly compounds of calcium, zinc, magnesium and phosphorus, which remains in the filter as ash and gradually takes up channel volume (Sappok and Wong, 2007). Ash raises back-pressure and shortens the interval between regenerations, and it can be removed only by taking the filter off the vehicle and cleaning it. The rate of accumulation follows oil consumption, and an ageing engine with worn rings, valve guides or turbocharger seals consumes more oil than it did when new. Engine condition is therefore the first determinant of maintenance burden, ahead of anything to do with the filter.

Catalysed systems increase the fraction of nitrogen oxides emitted directly as nitrogen dioxide. Carslaw (2005) linked a rise in that fraction at London roadsides partly to the fitting of catalysed filters to buses, which shows that a retrofit can improve one pollutant while worsening another. On durability, the evaluation of continuously regenerating filters on New York City transit buses by Lanni et al. (2001) remains an instructive published example of a fleet trial with ultra-low-sulphur fuel, emissions testing and service records taken together. Particle emissions also need to be measured appropriately: filters remove solid particles very effectively, but the choice between mass, number and opacity as a metric affects what is seen (Kittelson, 1998; Giechaskiel et al., 2014).

Each candidate bus is assessed before any filter is ordered. The assessment covers remaining service life, since a filter fitted to a vehicle due for withdrawal will not repay its cost; oil consumption from top-up records over a defined distance; free-acceleration smoke opacity; turbocharger and injector condition; and any history of coolant loss into the cylinders. Vehicles with high oil consumption or heavy smoke are repaired first or excluded. The purpose of the stage is to prevent the filter from being used to conceal an engine fault that will destroy it.

Exhaust temperature upstream of the proposed filter position is logged on representative vehicles, on their actual routes, for at least two working weeks. The record is reduced to the share of operating time spent above the temperature that the supplier specifies for passive regeneration. Urban bus cycles differ widely in speed, stop frequency and idling (Barlow et al., 2009), and a route with long idling and low speeds may never reach that share. Passive systems are assigned where the temperature profile supports them and active systems elsewhere. Fuel sulphur content and the lubricant specification are confirmed at this stage, because both affect catalyst performance and ash.

Before installation, the indicators in Table 1 are recorded for each vehicle over a period long enough to cover seasonal variation in fuel use, ideally a full year from existing depot records. A subset of vehicles is given an emissions measurement by whatever method is available to the operator, whether a chassis dynamometer, a portable system or an opacity meter, with the method documented so that it can be repeated unchanged.

Installation follows the supplier's instructions, with attention to the distance from the turbocharger, insulation of the pipework and the absence of leaks ahead of the filter. Every vehicle receives a back-pressure sensor and a temperature sensor connected to a logger and to a driver warning lamp. The clean-filter back-pressure at a defined engine speed is recorded as the reference against which later readings are judged. Drivers and workshop staff are briefed on what the warning means and on what must not be done, in particular continuing to operate with the lamp lit.

Logged back-pressure and temperature are downloaded at each routine service and plotted against distance. A rising trend that does not fall back after regeneration indicates ash and schedules a cleaning; a sudden rise indicates failed regeneration and prompts inspection of the engine and the route allocation; a sudden fall indicates a cracked or melted substrate. Maintenance on the filter system is booked to a dedicated job code so that its labour and parts can be separated from other work. Without that separation the maintenance burden cannot be measured at all.

At fixed intervals the indicators are compiled and compared with the baseline. Two points of method apply. First, a simple before-and-after comparison is vulnerable to changes that affect all buses, such as a new timetable or a change of fuel supplier. Where the retrofit is phased, vehicles not yet fitted serve as a comparison group over the same period, a design discussed by Shadish et al. (2002). Second, cleaning intervals and filter failures are time-to-event data, many of them incomplete at the time of review because the filter is still in service. They should be analysed with methods that handle such censoring (Meeker and Escobar, 1998), not by averaging the events that happen to have occurred. The review then informs maintenance intervals in the manner of reliability-centred maintenance (Moubray, 1997) and the screening rules for the next batch.