Current UK design guides suggest that HFDs should achieve an operational life of approximately 10 years. The filter material is then expected to require replacement or recycling, and this is often included in maintenance plans. However, during filter drain field evaluations (Farrar, 1994; Farrar and Samuel, 1989; Samuel and Farrar, 1988), acceptable performance of drains have been observed after 20 years of operation with minimal or no maintenance undertaken. These reports have also shown that there can be a differentiation between the service life of type A and type B (types A and B define different gradings of filter material with type A being finer) backfilled trenches and an implied correlation between filter specifications and modes of failure. Type A is reported to be prone to collecting road detritus at or near the surface of the drainage trench. Some of the key findings of these studies indicate poor construction standards and use of material that does not meet specification criteria, hence implying that the asset’s longevity may have been affected by poor construction quality. To fully capture a more accurate representation of HFD (average) service lives (and taking into account the aforementioned work), there is a need to further evaluate deteriorated drains and collect appropriate condition data.

Since these reports have been published, there has not been any other significant output (academic or industrial) whatsoever regarding the asset’s performance and/or its degradation characteristics and, more importantly, how to address key maintenance planning issues.

One exception to this, Rowlands and Ellis’s (2007) work, addresses in situ recycling as a means of introducing a level of sustainability and control over HFD maintenance operations. The authors focus on a novel rehabilitation approach and offer an overview of waste management issues during maintenance work but do not introduce a theoretical framework to support intervention time frames and holistic management; they instead adopt the existing empirical HFD service life projections often employed in the sector.

In the meantime, Highways England (HE) has brought forward a policy of non-recommendation for use of HFD in new projects, and this drainage option remains in UK standards for maintenance and rehabilitation operations. There are a number of reasons for this that may be attributed to design limitations (stone scattering as a safety hazard, cost of suitably graded stone in some regions), lack of engineered assessment methodologies to evaluate performance drop (malfunction of drainage system may go unnoticed) and the requirement for regular maintenance and replacement of filter stone at end of service life. These should presumably reinforce the case of introducing a holistic MS to deal with HFD in the roads network.

In terms of maintenance thinking, routine, corrective, preventive, predictive, proactive, and reactive strategies are terms often adopted in the industry to deal with the various approaches used for management of the physical assets in the UK’s highways. These are often interrelated; the most obvious distinction between the strategies is based on whether failure has occurred even though the definition of a failed state for a number of assets remains in many instances blurred (Uddin et al., 2013). It is thus easier (in terms of HFD management) to define maintenance using the terms proactive (prevent impending loss of acceptable drainage capacity and employ evaluation methods factorising condition or time) and reactive (run asset to failure and treat once drainage capacity is below acceptable thresholds, or a firefighting approach to maintenance (Swanson, 2001)). This of course requires a clear definition of the asset’s functions, how these are impaired, what the relevant failure modes are, what causes them and how these can be detected, and lastly what the impact on the pavement system and road user will be.

For HFD maintenance, interventions are still largely regarded as necessary repair work. While the importance of drainage has been highlighted as a main factor affecting long-term pavement performance (Cedergren, 1974; Hudson, 1968; Mallic and El-Korchi, 2013; Robinson et al., 1998) and driver safety (Johnson and Chang, 1984), there has not been a targeted effort to introduce the means for holistic drainage management or decision-aid toolkits. Drainage is not central in pavement management systems (PMSs), and engineered assessment techniques and rational maintenance strategies are in many cases omitted or considered to a lesser extent (Haas et al., 1994; Robinson et al., 1998). HFD are often in service until they fail in removing surface runoff from the pavement in an efficient manner, and there is still no way to predict or even classify functional failures. The alternative maintenance thinking (the first being in essence reactive) suggests implementing a time-based approach. Under this strategy, HFD maintenance and rehabilitation may be specified in a cyclic fashion (annual or biannual maintenance cycles or hard-time replacement at the end of the HFD design life) – see Figure 1.

Figure 1 Current maintenance thinking, intervention triggers and philosophies

Time-based maintenance (TBM) policies are usually derived on the premise of a bathtub curve (Klutke et al., 2003), which maps the gradual deterioration of the asset to increasing failure rates as cause and effect after a given service period (Ahmad and Kamaruddin, 2012; Yam et al., 2001). Under this scenario, maintenance, renewal and rehabilitation (M,R&R) tasks are scheduled at predetermined time or service intervals, and in general terms, time between successive interventions remains constant for asset categories with similar design characteristics. If these assets do not present a narrow failure distribution pattern that can be accurately predicted (Swanson, 2001), two scenarios will subsequently be presented: one, only part of the useful life of the asset will be utilised between interventions, or two, failures will not be addressed in a timely manner. Both scenarios will eventually incur additional costs to road operators and users by imposing unnecessary treatments or unexpected failures, with the latter imposing additionally a critical safety hazard. Safety implications (in terms of limiting water accumulation on the pavement surface and all relevant hazards) thus commonly dominate the planning process of maintenance operations under a reactive philosophy.

This suggests the reassessment of the existing HFD management thinking in order to establish how effectively maintenance interventions are planned and undertaken. From the studies and field evaluations reported above, it can be concluded that the degradation characteristics of in service HFD will not (restrictively) be a function of operational life (or the current definition of HFD service life is misleading). Takata et al. (2004) suggest that rates of deterioration for various asset types depend on operational and environmental conditions, and clearly, for filter drains, construction quality and maintenance history will also have a role to play. Initial projections of design lives of 10 years seem to be an underestimate, but functional failures that are related to the surface water removal capacity can appear within this period and will require attention. Under a TBM approach, M,R&R actions that are usually trench scarifying or vegetation control can be specified on annual/biannual cycles; aggregate replacement (HFD reconstruction) is due after 5 or 10 years of service life.

A generic framework that describes the requirements for holistic AM is proposed by Hassanain et al. (2003) using five key operations, which are

• identify assets,

• identify performance requirements,

• assess performance,

• plan maintenance,

• manage maintenance operations.

Similarly defined AM frameworks can be found in the literature; in Wittwer et al. (2002), condition assessment and trade-off analysis are included in the basic cycle of an AM framework (physical condition and asset performance as descriptors of the effectiveness of service delivery are often used hand in hand), whereas Dornan (2002) goes into further detail in asset renewal/replacement analysis methods (life-cycle cost, cost-effectiveness analysis or equivalent annual cost) and asset disposal policies. In practice, even though different definitions can be found, an AMS is the combination of engineering and business practices to support decision making at the strategic, network and project levels; in other words, it is a way of doing business by adopting the right procedures to achieve results cost effectively with the limitation of sparse resources (Cambridge Systematics et al., 2006; Godau, 1999).

Individual asset-specific MSs like PMSs were founded on this theoretical framework focusing on the key objective of cost-effective infrastructure maintenance and operation. PMSs are essentially the collection of all tools, technologies and processes that help managers make better decisions and manage their pavements more effectively. Their implementation since the late 1960s (Markow, 1995) came as a response to the shift from design-and-build operations to the repair-and-maintain mode. At this point, PMSs are abundantly available for use but other highway-related MSs have emerged to deal with bridges (bridge management systems), safety (safety management systems) and in-house maintenance operations (maintenance management systems) (Li and Sinha, 2004).

While all these systems address different asset categories and functions within the highway network, they integrate at their cores the same systems thinking and management principles (this work emphasises how drainage management has not yet embedded such an approach). They are fundamentally information driven, heavily dependent on databases (inventory, condition, historic maintenance, budgets, M,R&R options/impacts/costs) and employ a particular type of analysis that allows asset managers to fine-tune interventions, optimise asset life cycles and carry out investment decisions for preservation, expansion and operation of any given network. The data required for a MS are produced by monitoring and inventory activities, while modelling provides the tools for planning, trade-off analysis, ranking, and optimisation (Li et al., 1997). For PMS, pavement performance models adopt either deterministic (Abaza, 2004; Abaza et al., 2001; Wong et al., 2003) or probabilistic (Abaza et al., 2004; Golabi et al., 1982; Li et al., 1997) approaches, and DSTs that integrate prioritisation or optimisation models.

The recently published Highways Infrastructure Asset Management Guide (UKRLG, 2013) complements the ISO 55000 (BSI, 2014) series on the topic of AM in the UK. The latter supersedes the previous AM standards (PAS, 2008) and now offers an overview of AM, the requirements for an AM system and the guidelines for the application of ISO 55001. The ISO is a high-level strategic guide that introduces systems thinking and AM organisational precursors in a business-oriented manifesto that aims to standardise AM. A tactically oriented document, the UKRLG (2013) guide focuses on physical assets in highway networks, offering 14 recommendations to achieve an appropriate level of benefit from AM. These recommendations deal with various themes, including life cycle plans, performance management and monitoring, risk management and AM policy and strategy. The drainage-specific guidance document HMEP (2012) in turn focuses on establishing a cost-effective approach to managing drainage assets, based on building drainage-specific databases (inventory, condition, maintenance intervals, frequency of failures, maintenance requirements, etc.) and using this information to apply relevant AM principles. The guide highlights various strategic requirements to be developed and offers a number of concise recommendations (i.e. the requirement to address the causes of problems as opposed to symptoms and the requirement to use asset data to focus, support and inform maintenance activities mapped to overall AM objectives). It also proposes the adoption of proactive and prioritised programmes and the need to support these using asset condition data.

All these documents describe a ‘high-level’ framework to be developed but do not go into detail with regard to the inventory data to be collected, the condition metrics to be established, the performance indicators to be introduced or, lastly, the deterioration and maintenance optimisation models to be built. For a number of asset types (i.e. pavements) a wide spectrum of these parameters can be found in relevant AM or asset assessment literature. This will not hold true when drainage systems are to be evaluated; this simply highlights that a top-down approach through the available strategic and tactical publications will enable road administrators, operators and local councils to establish the required AM framework (or the AM know-how) but eventually the practical end of such a system will need to be addressed: what kind of data are to be collected, how are these to be used and in what sense will the outputs reinforce the case of AM and be of true value.

At the lowest AM tier, with a more operational focused document, Spink et al. (2014) approach the issue of managing drainage assets in the latest CIRIA publication that addresses some of the questions raised in the previous sections. The guide thus includes elements for inventory and condition data collection and evaluation and a proposed performance assessment regime. Being primarily focused on condition appraisal, the publication defines structural and service conditions as the two main parameters to be extracted through drainage condition surveys building upon visual evaluation. Five discreet condition bands in a one-fits-all assessment approach are included in the text.

While the proposed bands could be used to characterise individual HFD sections, the methodology embedded in the assessment procedure to reach them is rather generic. The deterioration characteristics of HFD are not considered, and the boundaries proposed to establish the discreet condition states are largely empirical; these are derived by rating the asset employing visual means rather than identifying and measuring the type and extent of the severity of fouling in an evaluated section. In fact, no framework for intrusive (or other non-destructive) evaluation is described in detail and thus a large portion of the information required to assess HFD is omitted (severity of sedimentation and extent within filter media tied to HFD failure modes). The condition evaluation adopting this approach would retrospectively address only one of the (two) HFD failure modes; that would be the reduction of surface runoff removal capacity (functional failure) due to surface crusting (failure mode), and thus, the duality of this particular drainage system is ignored.

Performance assessment in the guide is similarly based on visual surveys, but these can take place only when water can be found within the system (making assessments impossible during dry periods) or the combination of the two condition components identified (structural and service condition as a proxy for performance). The former approach is largely reactive as it suggests addressing the symptoms of HFD deterioration (reduced performance) as opposed to what causes this (sedimentation and reduced void space in trench) but can potentially allow the identification of flooding hotspots and be of value at the network level. The latter is again based only on evaluating the exposed filter media and thus is bound to provide limited information on the overall drainability of a section.

In simple terms, the major issue with this assessment approach is that the essence of fouling, its extent within the HFD trench and how that correlates to the drainability of a section are not explicitly or adequately considered in determining maintenance and renovation strategies. Since the sedimentation levels are not quantified, the actual capacity of the drain to remove water from the pavement system is not considered. In truth, a visual inspection-based condition-rating index can offer a quick evaluation methodology at the network level, it is though based on rating (rather than measuring) physical characteristics of the aggregate fill and bound to ignore some of the relevant information required to draw conclusive remarks for the current condition of the asset.

Also of interest is how local authorities currently approach the matter of managing drainage assets; adopting the UKRLG (2013) code of practice, condition standards are proposed at three domains here: safety, serviceability and sustainability. Maintenance plans produced by councils, which can be considered the tactical and/or operational side of the AM systems in place, suggest managing drainage assets adopting a risk or a fire-fighting approach (Bournemouth Borough Council, 2008; Middlesbrough Council, 2006; Suffolk County Council, 2008; Walsall Council, 2012).

Such plans will generally include guidelines for managing all relevant transportation assets; in the cases of carriageways, footways or cycle routes (and drawing from the Well Maintained Highways code of practice), a number of condition parameters (deflection, surface defects, skidding resistance, weed growth, slipperiness etc.) are to be measured and mapped against the core objectives of a highway maintenance regime for each domain. Drainage condition standards lack any such parameters or metrics. Instead, the three core objectives are streamlined against flooding risk, which is in principle a measure of performance rather than physical condition. This can be a facilitator of a reactive maintenance philosophy or of a preventive one that is primarily based on subjective indicators due to the lack of other condition metrics.

It is evident that our current thinking is lacking; in terms of planning, management of investments and understanding the factors that drive the deterioration of HFD, our strategies are largely underdeveloped and are primarily based on empirical methods (and thus tend to be reactive). This is due to a combination of factors; drainage has been neglected in the past, efforts to implement assessment techniques only partially addressed AM requirements, strategic guides offer fundamental (strategic) concepts lacking the ‘engineering-end’ of the management equation and empirical and/or time-based approaches offer limited opportunities to evaluate the physical condition of an asset and thus to collect relevant and specific HFD data.

Maintenance thinking has evolved in recent decades to meet the requirements for greater dependence on business and engineering principles, use of benefits and costs through what-if and life-cycle scenarios, rationalisation (and thus lesser dependence on generic standards) and, lastly, accountability of data and of performance projections. The sector’s thinking has long moved from considering maintenance as necessary repair work, and the definition of Robinson et al. (1998) of ‘maintenance as a management issue concerned with delivering a defined quality of service, activities and procedures, timing of interventions and resources of people and materials’ conclusively illustrates the fact.

The adoption of AM principles in the highways sector has been briefly described in the previous section. By identifying the gap between introducing the AM know-how and embedding the engineering prerequisites for this to be of value, the development of a set of HFD management building blocks is proposed. Looking to formulate an HFD-specific AM subsystem, the particular characteristics of the asset are to be considered and evaluated in a framework that will be in line with other existing asset-specific AM systems and thus should include

• a geo (or network)-referenced inventory database of HFD,

• condition data classifying deficiencies accountable for drainability levels,

• maintenance history,

• network-level HFD condition distributions,

• ageing/deterioration modelling and estimation of remaining service life,

• intervention analysis and network level program costs mapped to projected condition,

• optimisation of investments,

• project level analysis and tactical overview of maintenance requirements for candidate projects.

The eight elements described here address the five key operations defined by Hassanain et al. (2003), and will in principle allow asset managers to approach the task of managing HFD employing a rational set of decision-aid tools. Some of these themes look trivial and a fair amount of relevant information can be found scattered in the existing literature (e.g. drainage inventory and/or condition data collection in Spink et al., 2014) potentially offering some indications in respect to how to advance the framework. Focusing on HFD-specific elements that deal with quantifying the severity and extent of what drives the drop in serviceability levels and reflect the engineering orientation of the system, a suitable platform that deals with the ad hoc asset design features, how these change during service life and what impact this has on drainage capacity, needs to be defined.

Further expanding the proposed HFD management framework, the need to develop and introduce the methodology to support such elements as condition assessment, deterioration modelling, DSTs and optimisation of investments are identified. The following sections present the case of rational condition measuring that is considered the cornerstone of the aforementioned elements, and the tools that could eventually be developed using engineering assessment techniques.

By employing a physical condition classification regime, an engineered assessment system is introduced to quantify the extent and severity of fouling within an HFD section. The next step in formulating the HFD management framework is to use this particular assessment approach in formulating a better understanding of HFD degradation characteristics and a methodology to establish discreet condition bands for the drains. HFD failure modes are thus further evaluated in order to identify and introduce these bands (considering that available assessment criteria (Spink et al., 2014) underplay the duality of this particular drainage design) and to investigate how deterioration can be quantified and modelled.

Discreet condition states: defining the boundaries and embedding maintenance activities

With no available quantitative data to represent the deterioration progress of HFD in the highways network, Figure 4 represents a conceptualised definition of six discreet HFD condition bands along with the currently available renewal strategies that restore the condition of the drains from a downstream band (lower drainage capacity), to an upstream condition (higher drainage capacity). The four main states proposed here are excellent condition, fair condition, poor condition and very poor condition with two subcategories representing the top-down failure modes: fair crusted and poor crusted (EC, FC, PC, VPC, FCC and PCC, respectively).

Figure 4 General ageing/renewal model proposed for evaluation of HFD

The two crusted condition bands are critical for safety reasons and are introduced here to enable separation of the two identified failure modes through network surveys and previous field studies. Lastly, the very poor condition represents an HFD section that has surpassed its projected service life (reduced surface and subsurface drainage capacities) and which presents a foulant-dominated aggregate fill. The effect of the various available maintenance options on the overall condition of a section is also visualised in the diagram.

It is crucial to identify and establish for such a concept relevant criteria to form the boundaries between each condition state (linked to the condition assessment regime proposed). The field and laboratory data presented in Stylianides et al. (2015) suggest that a layer-by-layer analysis is required to classify the serviceability levels of the trench and to evaluate the different failure modes. The FC and PC states, for example, are manifested in the uppermost of the drain and can thus be easily identified under a reactive regime; if a more proactive (or on-condition) approach is required (and this paper suggests that this is in fact the case), a quantification of fouling levels (FL) mapped to anticipated drainage capacity should be achieved.

The permeability trials conducted exhibit how k _v as a function of FL can be extracted in a controlled experiment. Being a free draining material (void ratio of type B material measured to range between 0·66 and 0·80) with a large void space, a major drop in permeability values (and thus drainage capacity) is expected only when fouling extent reaches significant levels within the trench. A foulant-dominated fill will be one for which the drainage capacity will be a function of the foulant rather than the aggregate fill (k _v < 5 mm/s, R _FV < 0·6). Its functionality will thus be hindered and the removal of surface or subsurface water will be reduced. We can thus establish certain boundaries between discreet condition states, and these are presented (in a qualitative manner) in Table 1 linked to anticipated FL.

Table 1 HFD discreet condition bands and condition criteria and anticipated drainage performance

Table 1 HFD discreet condition bands and condition criteria and anticipated drainage performance

Condition band	Condition criteria		Anticipated performance
	Surface	Subsurface	Surface	Subsurface
1. EC	Highly porous (minimal to zero FL)	Highly porous (minimal to zero FL)	Free draining	Free draining
2. FC	Highly porous (minimal to moderate FL)	Highly porous (minimal to moderate FL)	No impact on drainage performance	No impact on drainage performance
3. FCC	Reduced porosity (moderate to high FL)	Highly porous (minimal to moderate FL)	Reduced water removal capacity	No impact on drainage performance
4. PC	Minimal porosity (high to very high FL)	Minimal porosity (high to very high FL)	Minimal water removal capacity	Minimal water removal capacity
5. PCC	Foulant-dominated fill	Minimal porosity (high to very high FL)	Zero water removal capacity	Minimal water removal capacity
6. VPC	Foulant-dominated fill	Foulant-dominated fill	Zero water removal capacity	Zero water removal capacity

Employed with an HFD generic ageing/renewal model, asset owners can assess and characterise filter drain sections within an evaluated network according to quantitative fouling and performance levels. In line with PMS presented in previous sections, to add value to this model, a mathematical representation of ageing and renewal policies should be considered.

In its simplest form, a generalised deterioration/renewal mathematical model conceptualised by De La Garza and Krueger (2007) is depicted in Equation 2. In principle, the model determines the annual change in the asset condition by calculating the effects of maintenance treatments on the various condition levels and then ageing the resulting condition distributions according to the adopted deterioration rates.

$$Conditio{n}_j(t+1)=Conditio{n}_j(t)-\frac{Conditio{n}_j(t)}{D_{jk}}+\frac{Conditio{n}_i(t)}{D_{ij}}+{\displaystyle \sum R_{kj}(t)-{\displaystyle \sum R_{ji}(t)}}$$

2

Deterioration rates (denoted by D _ij and D _jk ) can be extracted using either probabilistic or deterministic means, while R _kj and R _ji denote the upstream condition changes due to maintenance interventions (i.e. the impact of maintenance options on condition states). Focusing back to HFD management requirements, the impact of treatment options on condition states (R values) and a methodology to predict deterioration (D values) using the proposed condition classification framework (Table 1, Figure 4) should be defined.

An example of such a work in the sector has recently been presented by Costello et al. (2011), which describes a DST for managing ancillary assets using proactive principles. The methodology in this work draws from standard inventory collection practices and adopts five discreet condition bands and probabilistic Markov chains to simulate the deterioration of the assets (and to tackle the absence of historic condition data reported in the work). Once the required data for year one is collected (using a simplified CAS), a base condition vector is established to describe the condition distribution of assets within the network with bands ranging between excellent to very poor. The authors then calculate and present deteriorated states (the ageing process) using a transition probability matrix throughout the planning horizon in a given network. Having no records of deterioration rates, engineering judgement is used and a linear model to set up a life-cycle planning model is defined.

Looking back to HFD deterioration understanding, empirical evidence suggests a larger than 10-year operational life (reported to extend past the 20-year mark in some cases). Similar to the ancillary assets case, the typically reactive HFD management approach falls short in terms of enabling the collection and use of relevant condition information that could be embedded in a deterioration model (and thus allow the extraction of relevant D values as per De La Garza and Krueger, 2007). To enable a network level evaluation of HFD and predict the condition of the drainage network in discreet condition bands, a Markov probabilities transition matrix is proposed to represent ageing in a defined network. The six condition states (n-EC, FC, FCC, PC, PCC, VPC) introduced in this work (Table 1) are adopted, and an asset life (L) equal to 25 years is used; lastly, the deterioration is assumed to progress in a linear manner. The calibration and the level of accuracy of such a probability matrix will inevitably be a result of the collection of annual network-level condition data.

Using this scenario and adopting a base condition vector to denote 100% of the HFD network to lie within the excellent condition band (year 0), the progression of deterioration can be extracted for the whole network using the transition probabilities (extracted using Equations 3a, 3b). This is done by basic matrix manipulation; to extract the first year’s deteriorated vector ( a ₁), one needs to multiply the base condition vector ( a ₀ = [1, 0, 0, 0, 0, 0]) with the probability matrix P; to calculate the condition distribution at any future year, a _t = a _{t −1} P = a ₀ P ^t can be used.

$$\alpha =\frac{n-1}{L}$$

3a

$$P=\left[\begin{array}{cccc}{p}_{11}& p_{12}& \dots & p_{1n}\\ p_{21}& p_{22}& \cdots & p_{2n}\\ ⋮& \ddots & \ddots & ⋮\\ p_{n1}& p_{n2}& \cdots & p_{nn}\end{array}\right]=\left[\begin{array}{cccccc}1-2\alpha & \alpha & \alpha & 0& 0& 0\\ 0& 1-2\alpha & \alpha & \alpha & 0& 0\\ 0& 0& 1-\alpha & 0& \alpha & 0\\ 0& 0& 0& 1-2\alpha & \alpha & \alpha \\ 0& 0& 0& 0& 1-\alpha & \alpha \\ 0& 0& 0& 0& 0& 1\end{array}\right]$$

3b

Adopting the transition probabilities matrix (TPM) and using the parameters described in the previous paragraphs, one can graphically represent the deterioration of HFD in a given network as shown in Figure 5.

Figure 5 Distribution of HFD network condition bands adopting L = 25 and linear deterioration progression; a ₀ = [1, 0, 0, 0, 0, 0]