*6.1. Embankment Instability Metric Thresholds*

As suggested, in previous work to establish this technique (Sharpe and Hutchinson, 2015) [25], thresholds of geometry deterioration have been suggested based on observations from failure sites. The study presented in this paper examined a further 51 known failure sites; of these, there are 28 sites which have sufficient data in the reported time period of failure, and examination of the track geometry data suggests clear signs of failure in 19 of these cases. For these 19 sites, the maximum embankment instability metric values range between 4.2 mm/yr. and 12.8 mm/yr. for all years examined over all 10-yard sections of the failure sites.

This assessment of the rate of deterioration for these earthwork failures, combined with an understanding of the likely rates of deterioration due to trackbed failure and the effect of maintenance, confirm the assertion of the following suggested risk level thresholds below in Table 2. In this project, "risk" is defined as the effect of embankment problem on track system performance, with the purpose of highlighting the appropriate maintenance action to be undertaken. These thresholds apply to the calculated embankment instability metric and are the same values as shown in Sharpe's study.


**Table 2.** Embankment instability metric threshold descriptions.

It should be noted that these risk thresholds are only intended as guidance. Presently, there has been limited calibration of these values and they are based on limited observations from known failure sites. Although an average metric value for an embankment asset has been considered as a measure in this study, it is recommended that the risk classification thresholds apply to the in-year 10-yard metric values, rather than an averaged metric value.

The thresholds above have been used to show the split of observed embankment instability metric values for the population of all embankment assets analysed in this study. Considering the max metric value generated (for all years and 10-yard sections) of each embankment asset, the thresholds showed 4% negligible, 23% minor, 41% moderate and 32% high risk.

## *6.2. Recommended Track Geometry Recording Frequency Threshold*

For the results presented along with this project, a minimum of two track geometry recordings are required in any one deterioration year, to calculate an embankment instability metric value for that year in that section. However, the study has suggested that a higher frequency of recording is recommended to calculate metric values.

The numerical simulation used to model changes to track geometry recording frequency suggests minimum and recommended thresholds for the recording frequency in order to reliably calculate the embankment instability metric. The thresholds are shown in Figure 7, and it is suggested that the recommended thresholds are applied. These thresholds should be calculated by assuming a deterioration rate of 4 mm/yr. (as this is the high-risk level), for example a line speed of 110 mph would require a frequency of six records per year. that the recommended thresholds are applied. These thresholds should be calculated by assuming a deterioration rate of 4 mm/yr. (as this is the high-risk level), for example a line speed of 110 mph would require a frequency of six records per year.

*Infrastructures* **2019**, *4*, x FOR PEER REVIEW 14 of 18

### *6.3. Sensitivity Analysis 6.3. Sensitivity Analysis*

A sensitivity analysis has been conducted to consider the effect of other variables on the metric (such as track curvature, tonnage of rail traffic, etc.) and the metric has been shown to be independent to these other variables. A sensitivity analysis has been conducted to consider the effect of other variables on the metric (such as track curvature, tonnage of rail traffic, etc.) and the metric has been shown to be independent to these other variables.

Influence of track curvature: since the embankment instability metric considers the deterioration rate of alignment and differential Top in the calculation, it is worth considering whether the curvature of the track, directly related to the track cant, has significant influence on the deterioration of the alignment. The chart below (Figure 9) shows the variation of the embankment instability metric in comparison to the change in track cant. As can be seen, there is no evidence to suggest any correlation between the alignment deterioration rate or embankment instability metric with track curvature. Influence of track curvature: since the embankment instability metric considers the deterioration rate of alignment and differential Top in the calculation, it is worth considering whether the curvature of the track, directly related to the track cant, has significant influence on the deterioration of the alignment. The chart below (Figure 9) shows the variation of the embankment instability metric in comparison to the change in track cant. As can be seen, there is no evidence to suggest any correlation between the alignment deterioration rate or embankment instability metric with track curvature.

**Figure 9.** Graph of track curvature vs. embankment instability metric value. **Figure 9.** Graph of track curvature vs. embankment instability metric value.

Influence of tonnage: it has been suggested that the parameters being measured which form the embankment instability metric are mainly influenced by movements in the supporting embankments, rather than being affected by deterioration of the track and trackbed. The predominant factor influencing track and trackbed deterioration is the tonnage of the rail traffic. Therefore, it follows to consider any potential correlation between the tonnage and the embankment instability metric values. Figure 10 presents the values of the average embankment instability metric for increasing tonnage values (known as MGTPA, million gross tonnes per annum). It can be seen that there is some weak correlation of the tonnage with the embankment instability metric, suggesting that either (1) there is some effect which an increased tonnage has on instability of the embankments, possibly that increased tonnage may be exacerbating the rate at which the instability develops; or (2) that the embankment instability metric is influenced to a small extent by the general deterioration of the track and trackbed. It is not thought that this relationship is significant enough to consider that the embankment instability metric is invalidated or that tonnage needs to be taken into account. Influence of tonnage: it has been suggested that the parameters being measured which form the embankment instability metric are mainly influenced by movements in the supporting embankments, rather than being affected by deterioration of the track and trackbed. The predominant factor influencing track and trackbed deterioration is the tonnage of the rail traffic. Therefore, it follows to consider any potential correlation between the tonnage and the embankment instability metric values. Figure 10 presents the values of the average embankment instability metric for increasing tonnage values (known as MGTPA, million gross tonnes per annum). It can be seen that there is some weak correlation of the tonnage with the embankment instability metric, suggesting that either (1) there is some effect which an increased tonnage has on instability of the embankments, possibly that increased tonnage may be exacerbating the rate at which the instability develops; or (2) that the embankment instability metric is influenced to a small extent by the general deterioration of the track and trackbed. It is not thought that this relationship is significant enough to consider that the embankment instability metric is invalidated or that tonnage needs to be taken into account.

*Infrastructures* **2019**, *4*, x FOR PEER REVIEW 15 of 18

**Figure 10.** Graph showing variation in tonnage vs. embankment instability metric value. **Figure 10.** Graph showing variation in tonnage vs. embankment instability metric value. **Figure 10.** Graph showing variation in tonnage vs. embankment instability metric value.

Influence of line speed: consideration should be given to the effect of line speed on the embankment instability metric, which is displayed in Figure 11. As can be seen, there is negligible correlation between the line speed and the embankment instability metric value. Influence of line speed: consideration should be given to the effect of line speed on the embankment instability metric, which is displayed in Figure 11. As can be seen, there is negligible correlation between the line speed and the embankment instability metric value. Influence of line speed: consideration should be given to the effect of line speed on the embankment instability metric, which is displayed in Figure 11. As can be seen, there is negligible correlation between the line speed and the embankment instability metric value.

**Figure 11.** Graph showing variation in line speed vs. embankment instability metric value. **Figure 11.** Graph showing variation in line speed vs. embankment instability metric value.

### **Figure 11.** Graph showing variation in line speed vs. embankment instability metric value. **7. Discussion 7. Discussion**

**7. Discussion**  The earthworks that Network Rail manage pose a challenge on asset managers: aging The earthworks that Network Rail manage pose a challenge on asset managers: aging The earthworks that Network Rail manage pose a challenge on asset managers: aging infrastructure.

infrastructure. The majority of the British rail network was constructed largely before the development of modern geotechnical practice, and the modern network still runs on a foundation of the earthworks constructed before 1900. Moreover, the age of the NR earthwork assets also pre-dates detailed record infrastructure. The majority of the British rail network was constructed largely before the development of modern geotechnical practice, and the modern network still runs on a foundation of the earthworks constructed before 1900. Moreover, the age of the NR earthwork assets also pre-dates detailed record The majority of the British rail network was constructed largely before the development of modern geotechnical practice, and the modern network still runs on a foundation of the earthworks constructed before 1900. Moreover, the age of the NR earthwork assets also pre-dates detailed record keeping of stability interventions, undertaken on the earthworks in a form that is readily accessible today [3].

keeping of stability interventions, undertaken on the earthworks in a form that is readily accessible today [28]. Regulated industries, such as NR, are constantly challenged to demonstrate continuous improvement to their management processes. Today, the biggest aspiration is for a safe, reliable, efficient and sustainable infrastructure that is continually improving. This requires a well developed keeping of stability interventions, undertaken on the earthworks in a form that is readily accessible today [3]. Regulated industries, such as NR, are constantly challenged to demonstrate continuous improvement to their management processes. Today, the biggest aspiration is for a safe, reliable, efficient and sustainable infrastructure that is continually improving. This requires a well developed capability in asset management, with an appropriate and proportionate management of risk, whilst Regulated industries, such as NR, are constantly challenged to demonstrate continuous improvement to their management processes. Today, the biggest aspiration is for a safe, reliable, efficient and sustainable infrastructure that is continually improving. This requires a well developed capability in asset management, with an appropriate and proportionate management of risk, whilst recognising there is a degree of risk that is tolerable.

recognising there is a degree of risk that is tolerable.

recognising there is a degree of risk that is tolerable.

capability in asset management, with an appropriate and proportionate management of risk, whilst

Stopping trains from finding failed earthworks that have rapidly lost the ability to perform is one of the top geotechnical challenges. The majority of the potential earthwork instability sites identified to date are located on embankments [17].

The previous study [25,26], demonstrated that the deterioration of the trackbed performance can be measured through the observation of relevant track geometry parameters (vertical alignment (Top) and lateral alignment). Understanding these deterioration rates, their relationship to other trackbed data and the effect of track maintenance, can facilitate modelling of trackbed performance. In general, earthwork movements appear to be characterised by excessive deterioration in both lateral alignment and difference in top. Both parameters were evident in known areas of earthwork failure.

We have to continually develop better technology at reducing cost and steadily evolve more efficient methods for moderating risks.

This study establishes an improved methodology to understand how earthwork movements will appear in the processed data, both in the time-history charts and the embankment instability metric which is derived from them. Failures often show movements for many years before becoming critical (i.e., before the earthwork movement starts to cause track geometry problems that cannot be rectified within a practicable planned maintenance cycle). The early signs of imminent failure are therefore evident in the track geometry data, although prediction of failure is difficult as an increase in rate of movement may simply be a symptom of erratic behaviour. The data can be affected by both the seasonal variation of earthwork movements and the effects of earthwork problems on the track geometry recording process.

Deterioration of the track geometry related to earthwork movements can be captured by the track geometry data, provided that the recording frequency exceeds the recommended threshold, and the rate of deterioration is not substantially above 4 mm/yr. It is important to note, that if the actual rate of deterioration is substantially above 4 mm/yr. and the track geometry recording frequency is not high enough, the track geometry data may appear as erratic and the reliability of estimating the rate of deterioration is reduced. The deterioration rate and recording frequencies at which the data may become erratic will vary with the line speed. The recommended thresholds in Figure 7 should be used as a guide to understand the required recording frequencies to give reliable estimates of the actual rate of deterioration, and hence embankment instability metric values. Further work is required to determine the reliability of the metric when the deterioration rates and recording frequencies do not meet the recommended thresholds.

As suggested in previous work [25] to establish this technique, and explained in Section 4.2.1, thresholds of geometry deterioration have been suggested based on observations from failure sites. This study has examined a further 51 known failure sites; of these, there are 28 sites which have enough data in the reported time period of failure and examination of the raw data suggests clear signs of failure in 19 of these cases. For these 19 sites, the maximum embankment instability metric values range between 4.2 mm/yr. and 12.8 mm/yr., for all years examined over all 10-yard sections of the failure sites.

This assessment of the rate of deterioration for these earthworks' failures, combined with an understanding of the likely rates of deterioration due to trackbed failure and the effect of maintenance, confirms the assertion of the suggested risk level thresholds in Table 1. These thresholds apply to the calculated embankment instability metric and are the same values as shown in Section 4.2.1.

It should be noted that more calibration of these values is needed, as future work and a wider study to understand the variance of the embankment instability metric for embankment assets with no known history of instability would help, giving a context to the proportion of risk threshold breaches.

### **8. Conclusions**

The project presented in this paper demonstrates that track geometry data are a viable source to consider for detection of railway embankment instability.

Thanks to the development of an algorithm, a value of EIM, and so a measure of the asset vulnerability to failure, was assigned for each 10-yards (almost 10 meters) of track for embankment assets. Analysing a sample of 51 known failure sites, the EIM clearly showed evidence of high track geometry deterioration, consistent with failure.

The frequency of track geometry data recoding is an important consideration and data availability is a prerequisite for reliable analysis. Data coverage is one major limitation of this technique; typically only a quarter of the network has sufficient data to analyse the past three sequential years of earthwork performance.

A sensitive analysis was conducted to consider the effect of other variables on the metric and the metric was shown to be independent to those other variables.

As a logical extension to this project and algorithm developed during the study, further works are suggested. The current alignment process (shifting and rubber-banding) of the track geometry data would be extremely improved with the introduction of an automatic process. This will reduce the time and cost required to process and trend data to calculate the EIM and visualize the data for interpretation.

User input is also required to identify erroneous track geometry data. Automated detection and purging erroneous data would be a logical extension to this project, although should be solved within Network Rail systems, possibly making use of machine learning to improve the efficiency of the task.

The work completed in this study has been focussed on the analysis of embankment assets identified as in need of remedial work. Therefore, there is no global reference distribution from this analysis to demonstrate how the embankment instability metric performs for other embankment assets with no known history of instability. Such a study will assist with understanding what level of false positives may be generated through scaling up this analysis, and help to quantify other factors which may be influencing the metric.

**Author Contributions:** Writing—original draft preparation, G.S.; writing—review and editing, D.K., M.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** The research presented in this paper was carried out as part of the H2020-MSCA-ETN-2016. This project has received funding from the European Union's H2020 Programme for research, technological development and demonstration, under grant agreement number 721493.

**Conflicts of Interest:** "The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
