Based in on the edge of the Marches, iSpaniel Ltd provides robust and affordable container tracking, tracing and management solutions for brewers and other industry sectors. They specialise in tracking beer barrels and related items from delivery and collection to pubs and other outlets.
iSpaniel Ltd uses robust Radio Frequency (RFID) tags and Near Field Communication (NFC) to uniquely identify the combination of an individual container location, its contents and collection profile in real-time. Their SaaS (Software as a Service) cloud- and smartphone-based system provides customers with data analytics and alerts based on this detailed raw data.
iSpaniel Ltd wanted to understand the business potential for increased efficiency and sales volume by using existing data more effectively and combining it with external data to derive novel insights.
They hoped that by combining iSpaniel Ltd’s geospatial data relating to asset locations over time, with external data sources such as weather, demographics and events, they could more precisely predict sales and the readiness of assets for collection.
A further aim was to apply this business intelligence to optimise the routes taken by delivery vehicles and incorporate pre-emptive asset collections into these routes, reducing travel time and cost and minimising the time during which assets sat idle and empty.
iSpaniel Ltd’s central goal was to identify relevant data sources and explore algorithms to combine these sources with their own historical asset location data in novel ways, allowing them to assess the feasibility of predicting what asset would be required at a given location at a specified time.
This would be an achievable substitute for predicting sales, and would help iSpaniel Ltd to more effectively meet their customers’ requirements. They also hoped to use historical delivery data and future delivery schedules, along with information on routes and vehicle loading weight and delivery time constraints, to provide algorithms for optimising delivery schedules.
Think Beyond Data carried out an extensive research analysis into applying machine learning techniques to iSpaniel Ltd’s asset location data, in combination with various sources of external data. These techniques included Artificial Neural Networks, Logistic Regression, Decision Trees, Random Forests, Self-Organising Maps and Support Vector Machines.
Algorithms were trained on large amounts of asset location data, in combination with weather data at various levels of granularity, and postcode-level demographics. The team investigated both freely-available and subscription data sources, and while the initial investigation showed no definitive correlations with sales, iSpaniel Ltd found the exercise useful for future planning.
The second aspect of the project was to demonstrate how optimisation could help brewery businesses to reduce costs and CO2 emissions by scheduling deliveries more effectively. iSpaniel Ltd provided data describing schedules for two years’ worth of beer deliveries, in which each barrel was tracked by scanning an RFID tag to provide a GPS location.
Seven vehicles were available, ranging from a small van to a large lorry with differing weight limits. The large lorry also had a tachometer requiring the driver to take 45-minute breaks every 4.5 hours. These constraints added to the complexity of the challenge, since each vehicle had an individual cost per mile associated with it.
The Think Beyond Data team implemented a bespoke heuristic optimisation approach known as Ant Colony Optimisation to improve delivery schedules. Ants, in nature, successfully navigate a landscape in an optimal way by using pheromone deposits to communicate the directions that ants should take. This approach can also be applied to van delivery problems. Based on the data supplied, the team were able to successfully demonstrate how deliveries could be optimised for any given day to minimise costs.
They also increased the complexity of the problem by considering larger challenges across multi-day scenarios of up to a month, and combined the daily delivery schedules into problems consisting of a week, a fortnight or a month of deliveries.
This enabled the team to compare optimisation day by day, or longer term as required. The results from these experiments are shown in Figure 1 for daily, weekly, fortnightly and monthly optimisation.
Figure 1: An illustration of longer term, larger scale fleet optimisation for better coordinated deliveries and the benefits that can be realised by the brewery company from reduced fuel costs
iSpaniel Ltd found the feedback to be very useful in terms of guiding future research and product development planning. A comprehensive report was produced to summarise the project’s results, main findings and recommendations for further research.
The results clearly demonstrated that increasing the size of the optimisation problem from daily deliveries to up to a month can result in significant fuel cost reductions of up to 20%, or £1,250 per month, as well as a reduction in CO2 emissions of 3.57kg per vehicle per day.
These results illustrate the ability of Think Beyond Data’s bespoke algorithms to successfully scale to fleets of up to 140 vehicles of differing types and capabilities, and iSpaniel Ltd now plan to use this detailed data to integrate route planning into their systems.