Downer is using analytics to plan which road faults to prioritise repairing, which crews to dispatch and the most efficient route to drive.
Elton Shi (second from the right) and David Ming (third from the right) accepting the Centre for Business Analytics’ 2023 Practice Prize.
The engineering company was awarded the Centre for Business Analytics’ 2023 Practice Prize on Wednesday for the in-house built, automated work packaging solution.
Downer data and analytics manager for optimisation David Ming said at the awards night in Melbourne that Downer’s subsidiary DM Roads had configured the scalable planning system across all its Australian maintenance contracts, covering more than 28,000 kilometres of roads.
Paving the fastest route
Manually planning which defects to prioritise “from a database of potentially thousands” had previously been a huge challenge, Ming said.
Even though Victoria has a “dedicated planning centre” for its five depots, it still took “about two-to-three hours every day to craft one of these plans.”
“There could be important defects like on a highway; that requires priority; we need to do those first,” Ming said.
“If we were to only work on high-priority defects – say we only did potholes – we’d end up zigzagging across the state, and we’d probably end up spending more time driving than actually doing meaningful work, and if something new comes up that plan’s no longer valid.”
Ming added another approach – allocating crews to cover “very localised areas” – would also be inefficient.
“Where your crew’s concentrating might sometimes have much less important defects than other areas – and all those crews are subject to capacity constraints and weather emergencies.”
Ming said that Downer needed an automated planning capability that “does more work with less time and less resources”; and it made more sense to build it in-house because of the specialised knowledge required.
“We do use partners for some of our problems, but with this problem – it’s very difficult to retrofit a solution to our domain. And contracting it out can also be quite costly…So we wanted to use analytics to amplify our domain knowledge.”
Mathematical optimisation and decision-making
Downer strategic improvements analyst Elton Shi said the automated, planning system is based on a mathematical optimisation model.
The system needed to use outcome-based criteria to select the most efficient decisions from all available options within defined parameters.
Shi said that the parameters could include “business constraints such as crew capabilities, and maybe an eight-hour shift time.”
The system’s three decisions were “what jobs you’re going to, who you’re going to allocate [crews] to and in what sequence you’re going to do them.
“The sequence is very important for efficiency; going from point A to B to C might be a lot more efficient than going from C to A.”
Shi said the criteria is needed “to make the decisions with the most optimal or efficient outcomes” because, “it’s not exactly immediately clear which set of decisions is better than the other.”
“We need a measurable quantity to quantify the quality of the solution. And we do that through a scoring function.”
The scoring function has “two main components”: “the job score – the priority and urgency of that job – so higher priority jobs are worth more,” and “the resources used to do those jobs; that could mean costs like a penalty for missing a job.”
A plan that completes six jobs in one day might score 496 whereas a plan that completes seven jobs and finds shortcuts to drive six fewer kilometres might score higher.
Shi added that planning leads could also tweak and customise the parameters of the model to meet fluid or specialised circumstances.
“The final lead might say, ‘We want to drive less; the optimal is very dynamic and changes with the parameters of the model.’”
Ming added that this was an example of the optimisation framework’s scalability because “we can configure it with different contracts.”
Ming said that the capability’s main benefit to DM Roads is that it helped fulfil the Downer subsidiary’s purpose – making transport safer – but another is how much more accurate it made performance predictions.
“And that’s great for having constructive conversations with both our crews and our customers,” he said,
“Now, it’s based on, not just opinion, but a principled method.”