To address the current limitations of the ALCAM process, NTRO was engaged in 2022 to lead and deliver a collaborative multi-stage project funded by the Department of Transport and Planning (DTP) Victoria. This project sought to investigate how the ALCAM process could be optimised via the integration of modern technologies.
To evaluate the safety risks at Australian railway level crossings, the Australian Level Crossing Assessment Model (ALCAM) is employed. Traditionally, ALCAM assessments utilise information obtained through manual inspections undertaken by authorised personnel. As such, the process can be time-consuming, resource-intensive and limited by funding and inspector capacity/availability. This can result in level crossings being inspected inadequately (e.g. once every seven years), limiting the availability of up-to-date data for risk management and informed decision-making.
To address the current limitations of the ALCAM process, NTRO was engaged in 2022 to lead and deliver a collaborative multi-stage project funded by the Department of Transport and Planning (DTP) Victoria. This project sought to investigate how the ALCAM process could be optimised via the integration of modern technologies (see Figure 1), including automated data collection, virtual inspection techniques and machine learning. Delivery of this project would reduce resource-dependency and facilitate more frequent delivery of inspection data, making up-to-date and accurate inspection data available to stakeholders to utilise in managing risks and prioritising safety risk at level crossings.
The work to improve the ALCAM process was scoped as a three-stage process:
Stage 1 of the project was undertaken in 2022/2023 and focused on a desktop assessment of the current ALCAM approach. The existing ALCAM process was assessed to identify any potential untapped efficiency gains (e.g. optional alternative data sources to mitigate the need for on-site assessment data collection). Applicable technologies were also identified and investigated to understand associated costs and potential for use.
Outcomes of the literature review in this stage revealed the following:
Following the literature review, stakeholder and industry consultation was undertaken to identify and prioritise which ALCAM characteristics and LXM input parameters are time-consuming/difficult to assess on-site and would benefit greatly from automation. Six such characteristics (which correlate to 23 primary LXM input parameters and a further 100 secondary LXM input parameters) were identified as follows:
To facilitate automation of calculating the above parameters, technology currently utilised in the rail and road sector was reviewed in consultation with industry experts in data collection also including visualisation, software development and AI model development. The outcomes of this investigation indicated the following technologies (generally in various combinations to allow for calculation of ALCAM characteristics) are most applicable to this project:
The applicability of these identified technologies, in various combination (examples given in Table 1), were assessed to then determine the viability of undertaking AI model proof-of-concept trials in Stage 2 of the project. This assessment yielded a decision to develop AI models using LiDAR & video data from NTRO’s iScan vehicle to assess sight distance, signage/markings and clearance widths.
Stage 2 of the project was undertaken in 2023/2024 with the aim of developing and implementing AI-based models for feasibility assessment to optimise ALCAM processes. This stage of the project involved data collection, manual assessment, virtual assessment and the development and refinement of AI models.
Within this stage, ten Victorian locations, covering a diverse range of control types, mechanisms and traffic layouts, were identified to be surveyed and assessed. These locations were subject to both manual inspection and NTRO iSCAN vehicle (video and LiDAR) data collection methods, to gather information reflective of current industry practice and for the potential automated practices. This information was then supplied to collaborating partners to begin development of AI models for selected ALCAM/LXM parameter inputs. The development of the AI models broadly included:
In developing these models, a couple of challenging aspects were identified and addressed to ensure quality and accuracy of the resultant models and their outputs. These challenges were most notably:
At the end of Stage 2, the process for developing a digital twin framework for the ALCAM process was refined by evaluating the feasibility of automation through assessing the developed AI models and identifying attributes that require more detailed analysis. Continued refinement of these models was undertaken in Stage 3.
To improve consistency, efficiency and safety in support of reducing incidents at level crossings, Stage 3 of the project (2024/2025) undertook testing of integrating the digital twin model, as well as formulating a roadmap towards full automation of the field survey process. The focus on full automation in this stage of the project set out to:
This continued investigation into the feasibility of modernising and automating the assessment process for railway level crossings in Australia assessed the practicality, effectiveness and efficiency of the proposed solutions/interventions using a structured methodology as outlined in Figure 2.
The ALCAM evaluation in Stage 3 focused on refining the automation feasibility of attributes identified in earlier phases, as well as broader attribute categories applicable across different models. It considered attributes requiring design data, subjective interpretation or those currently unattainable due to data gaps. Feasibility testing at selected sites assessed the practicality of automating data collection, particularly for challenging attributes, using a mix of field studies, simulations and stakeholder input. Both qualitative and quantitative methods were applied to identify risks, limitations and improvement opportunities. Each attribute was evaluated across road and pedestrian level-crossing models, for their relevance, data needs and automation potential to streamline future ALCAM assessments. A total of 222 attributes were reviewed and categorised as follows:
Over the course of Stage 3, the methodologies applied to achieve meaningful outcomes from the feasibility testing covered the following steps:
By the end of stage 3, the key findings were noted as follows:
Overall, this project represents a major advancement in automating and modernising ALCAM, demonstrating through feasibility testing that technologies such as LiDAR, computer vision and digital twins can significantly enhance safety assessments at level crossings. However, further work can be done to continue advancing ALCAM automation, such as:
In addition to these recommendations, the project also establishes a foundation for an Accelerated and Automated ALCAM (A-ALCAM) model, where the term "acceleration" refers to approaches that enhance the accessibility and availability of ALCAM surveys, including the use of semi-automated and automated methods leveraging digital twin technology, machine learning and AI systems. As such, the project presents a strong case for future development focusing on:
