On 21 October 2024, almost three years after the collision between two passenger trains at Salisbury Tunnel Junction, two trains collided head-on near Talerddig, Powys, Wales. At first sight, the accidents were similar. Both involved poor adhesion, both involved a train travelling beyond its permitted movement authority, and both involved almost identical Class 158 or Class 159 vehicles.
However, as the 50,000-word RAIB report shows, other factors were involved at Talerdigg for which lessons need to be learned, not least in how ETCS is designed, how low adhesion is dealt with in ETCS, and how human factors impact on how drivers use ETCS information. This article summarises what happened and what might have been done to avoid the collision and how factors such as the characteristics of the site and ETCS were underlying issues.
The accident happened on the west side of a passing loop at Talerddig. Train 1J25, travelling from Shrewsbury towards Aberystwyth, was intended to stop in Talerddig loop so that train 1S71, travelling in the opposite direction, could pass. Both trains were two-car Transport for Wales (TfW) Class 158 diesel trains.
Instead of stopping, 1J25 passed its authorised stopping point, ran through the loop and entered the occupied single-line section beyond, where, at approximately 19:26, the two trains collided at a closing speed of approximately 50km/h. One passenger sadly died, four people were seriously injured, and 23 suffered minor injuries.
Investigation
RAIB’s investigation found that the immediate cause of the accident was train 1J25 passing its limit of movement authority (the ETCS equivalent of a red signal). This was due to three factors:
First, the wheel-rail adhesion in the area approaching the Talerddig loop was low (although not unusual for this area in October).
Second, the two sanding systems fitted to train 1J25 which might have enabled the train to stop within the loop and avoided the accident, did not dispense sand. The automatic sander failed to operate, and the manually operated emergency sander was not activated by the driver.
Third was that, given the low adhesion, the speed of train 1J25 approaching the entry to Talerddig loop was too high for the train to be able to be slowed sufficiently.
Having passed its intended stopping position, train 1J25 entered the single line beyond the loop. The adhesion on this steep downhill gradient was exceptionally low and even though train 1J25’s emergency brakes were applied, it did not decelerate as it approached train 1S71.
Poor adhesion in autumn conditions on rural lines is a known risk for railways that has been extensively discussed and is not repeated here. There was evidence that Network Rail had routinely operated its water jetting trains on the line and was aware of the risks. At Talerddig, the loop is on more or less level track. Train 1J25 was approaching from the east on a 1:80 up gradient, whereas train 1S71 was approaching from the west on a 1:56 up gradient. 1S71 was reported to be struggling with traction on the steep up gradient (only 50% axles powered).
This area was designated a high risk of low adhesion area (HRLA), and a traction gel applicator (TCA) had been fitted on this gradient, although RAIB found that it was not working. RAIB’s description suggests that the TCS had been fitted to assist trains working up the hill towards the loop where support for traction might be required and trains might be expected to stop. It would have seemed less likely that trains leaving the loop and entering the single track would need to stop on the down gradient.
Sanders
RAIB reported that 1J25’s brakes worked as intended but that the sanders did not. Soon after the accident, RAIB reported that the sander hoses on its leading car were blocked. However, the final report showed that this was a consequence not a cause and that if the sanders had been working normally, the compressed air used to dispense sand would have dislodged the blockage. There was no sand behind the blocked hoses demonstrating that sand discharge had not been attempted since the hoses became blocked.
RAIB found two issues: a high resistance contact in the Sander Isolation Switch and failure of Low Speed Relay to energise. Either failure would prevent the sanding system from operating. Moreover, although there is a test switch to check that the sanders work, it does not test these devices.
The driver operated emergency sander was also not used. This is a one-shot device that can only operate when the emergency brakes are applied. RAIB reported:
“The driver stated that it had not occurred to them to use the emergency sanders on the day of the accident. They recalled thinking that the adhesion conditions would improve, as this was their previous experience of braking in low adhesion conditions. The driver’s initial belief was that the train’s speed would reduce, in accordance with their expectation of how the train would respond in the situation. This expectation was based on previous experience which had resulted in successful outcomes.
“People update their assessments and their actions in an evolving situation. Their understanding of a situation may change, and their current understanding will, in turn, affect their attentional focus. Once they made the decision to call the signaller, the driver reported that their attention became focused on this task, and using the emergency sanders at this point did not come to mind.”
RAIB observed that the driver had never used the emergency sanders before and although the circumstances in which they were to be used were set out in the Rule Book and TfW’s instructions, applying those instructions in a stressful situation might not come naturally without training or practice.
RAIB’s simulations showed that if the automatic sanders had been functioning or if the emergency sander had been deployed when the train moved beyond its movement authority the collision would have been avoided.
Approach speed
The approach speed of train 1J25 was also a causal factor. The speed at which it approached the eastern entry to Talerddig loop meant that the required deceleration could not be sustained with the available adhesion. RAIB did not criticise the driver. It made the point that the driver was driving to the speeds shown on the ETCS Driver Machine Interface (DMI). It was the system that allowed a situation in which the margin between routine operation and unrecoverable overrun was too narrow when adhesion degraded and sanding was unavailable.
The driver of 1J25 called the signaller at Machynlleth to report that the train was sliding. The driver called on the normal channel, not an emergency call in the belief, from experience, that the train would stop soon after passing Block Marker MH1078. An emergency call would have alerted the driver of train 1S71 to stop and the signaller to take similar action for all trains. The signaller had taken 1J25’s driver’s calm communication as a sign that the situation was not serious. Although the signaller called 1S71, immediately after concluding the 1J25 conversation, there was no instruction to stop, nor was the signaller’s ETCS emergency stop feature used. In these cases, the collision would not have been prevented, but the consequences might have been reduced.
ETCS
An important difference from the Salisbury accident is the signalling system. Talerddig is on the Cambrian line which saw the UK’s first ETCS level 2 installation with no signals. All commands to the driver are provided by the DMI screen in the cab. As has been explained before in Rail Engineer, the system supervises train movement and can intervene with braking if a train is travelling too fast or is at risk of exceeding its movement authority (ETCS equivalent of passing a signal at danger).
In this accident, ETCS functioned as expected. It demanded full service braking in response to exceeding a speed limit and then emergency braking when the full service brake was ineffective. The problem was that ETCS assumed that the braking would be effective; it took no account of poor adhesion.
RAIB’s report discusses the issues with ETCS in poor adhesion including standards, practical implementation, and TfW’s knowledge base. It describes this as a possible underlying factor arising from a “longstanding misalignment between the safety model for the Cambrian ETCS signalling system and the safety model for the class 158 rolling stock”.
The standards in force at the time required trains’ braking performance and potential to pass beyond a limit of movement authority to be assessed using a “normal adhesion level” (not defined in standards of the time) and referred to “low adhesion” (also not defined) as out of scope. A design document from 2010 stated:
“Due to the impossibilities of accounting for all adhesion levels without making the system overly conservative in normal operation, very low adhesion levels would require a driver to adapt their driving style to control the speed of the train safely.”
Other standards also exclude poor adhesion from scope.
In reality, ETCS is fundamentally similar to all signalling systems which must be laid out based on some assumption of trains’ braking performance. Conventional signalling is generally laid out with a level of conservatism because supervision is intermittent. Because ETCS provided continuous supervision, its designers allowed reduced safety margins compared with more traditional arrangements. For example, the standard overlap beyond a signal is 180 metres whereas for Block Marker MH1078, the overlap was less than half at just 81 metres. Thus, it was even more important for the train to stop in the designed distance, but this was not recognised by the organisations involved.
Professional Driving Policies
The industry has recognised that the implementation of ETCS affects operators’ Professional Driving Policies which, alongside the introduction of TPWS have done so much to mitigate the risks of passing signals at danger and the consequences of so doing. In February 2026, RSSB released a 55-page report titled “How do professional driving policies for passenger operators change under ETCS? Understanding the ETCS driving task and the operational risks that are controlled either by the system or the driver.”
Operational factors
A further possible underlying factor was an incomplete understanding of how drivers interacted with the Cambrian signalling system.
Transport for Wales (TfW) seems to have believed that drivers were driving to professional standards, varying braking depending on conditions. However, TfW might have better understood how drivers were using the speed displays if it had known the frequency that the ETCS was intervening by applying the service brake in response to small exceedances of the indicated permitted speed, known as Service Brake Interventions (SBI).
SBI events are recorded in the ETCS Juridicial recorder (JRU). Cambrian Class 158 trains have two data recorders: the JRU and a conventional train data recorder known as the OTMR. TfW driver managers only had access to the OTMR and had to request Network Rail to supply information from the JRUs.
The difficulty of obtaining JRU information led TfW managers to review driver performance using only OTMR data. RAIB, with access to both data sources found that SBI events were happening much more frequently than TfW had believed, indicating that drivers were driving to the ETCS profiles i.e., focussing on the DMI and so tending not to take account of adhesion. DMI speeds cannot be varied for low adhesion as this functionality was not included in the Cambrian specification.
The report also observed that emergency preparedness for drivers (e.g., what to do if a train is likely to exceed its movement authority) could be improved, noting that the ETCS driving simulator built when the system was commissioned had fallen into disuse.
The nature of the DMI, how indications change colour quite subtly, and that up-coming changes in speed can be masked by the current movement authority section were also discussed in some detail with RAIB suggesting that the approach speed to the loop was simply the result of following ETCS prompts. This shows that the system behaviour is not fully understood by the operator, and training, monitoring, and risk assessment are likely to lag behind operational reality.
Recommendations
Building on its recommendations from previous adhesion incidents and collisions, including the most recent Salisbury collision, RAIB made nine recommendations and proposed one learning point. These are aimed variously at RSSB, Network Rail, TfW, train owners, and other passenger operators:
- Review and update the standards and guidance relevant to the design, maintenance, and testing of trainborne sanding equipment.
- Review of the design and maintenance of the automatic sanding systems fitted to class 158 units and where appropriate implement changes.
- Review the assumptions which formed the basis for the decision to undertake a simple overrun risk assessment on the Cambrian lines.
- Consider the opportunity to protect trains approaching the location of a potential exceedance of movement authority at the earliest opportunity as part of the specification for further ETCS schemes.
- Learn lessons from issues with the driver interaction with the DMI in this incident.
- Improved in-emergency-situation training for drivers.
- Continue the programme of research and testing to develop a more comprehensive understanding of the effectiveness and longevity of railhead treatments.
- Review structural standards for train interior fittings and when those standards should be applied to existing trains.
- Review the competencies required of staff whose duties require them to regularly work on trains and interact with passengers, including staff undertaking non-safety-critical duties.
Learning point: The accident underlined the importance of those with lead responsibility for safety critical conversations establishing a clear understanding with drivers so that appropriate actions can be taken to ensure the safety of the line and train operations.
Conclusion
The Talerddig accident shows how modern railway safety depends on the performance of a tightly coupled system of infrastructure, rolling stock, signalling, maintenance, training, and operations. If anything, ETCS makes system coupling even tighter. ETCS enables closer supervision which has led to reduced margins, which in turn increased reliance on predictable braking and effective sanding.
Clearly, slippery rails and a failed sander caused the Talerddig accident. But it has also highlighted that safety-critical systems must be assessed in an integrated way, including degraded modes. Assumptions about braking, adhesion and driver response must be tested against real operational variability. Assurance must prove that mitigations will work when demanded, not just that they can work during a test. Above all, the introduction of ETCS should not obscure the need to understand the physical limits of trains operating in adverse conditions.