Ready-to-use software (win-x64)
Digitalisation has made key information such as sound/image/video training samples, geographic conectivity (Open Street Map), weather information, etc. widely available to the public. With advanced instrumentation, companies can collect real-time information such as machine condition instantly. To maximise the benefits of digitalisation and turn information into savings, our products are rich in features and span across fleet optimisation, assignment optimisation, anomaly detection and data analytics. Besides traditional machine learning, we have hybrid algorithms integrating deep learning with optimisation
Color: route of control forecast. Gray: route of ensemble forecast
Ship Route Optimisation
IMO resolution A.528(13) (International Maritime Organization, 1983) advises ships to make use of weather information for routing. Our software is a mathematical approach utilising AI weather forecast to minimise fuel consumption and CO2 emission. By incorporating own fuel curve, ship operators can customise routing to individual ship to analyse performance and conduct what-if analysis without actual sailing. Its key features are as follows
Avoids unfavourable wind to find paths with good fuel efficiency and emission for ensemble of weather forecast
Find alternative routes for difficult weather and fine-tunes departure time
Multi-faceted and consider factors like waves and forbidden lat/lon due to geopolitical consideration
DIMS software
Fleet optimisation
Demand Integrated Maintenance Scheduling (DIMS)
Minimise fleet size for systems on preventive maintenance arising from demanded usage. Compute vehicle the spare ratio described in page 40 (Suggestions for further research) of 'Transit Cooperative Research Program 109 - System-Specific Spare Bus Ratios Update'
Assigning Systems On Round Trips (ASORT)
Minimise fleet size for systems on round trips, such as ferry, coach, etc.
Dynamic Shift Pattern Optimizer (DSPO)
Shift patterns of https://en.wikipedia.org/wiki/Shift_plan are efficient if workload remains the same for all shifts. Opportunity for saving arises when demands vary across shifts, for example, customer arrivals at 24-hour service centre are different for day versus night. For dynamic demands and rest requirements, DSPO computes efficient shift patterns using minimal staff size
Constrained clustering
Assignment optimisation
Constrained clustering
Assign points evenly to clusters with guarantee on attribute such as performance/similarity/distance. Applications are
Assignment of locations to autonomous clusters. In an efficient distribution system, supply points needs to be close to customers. Given a series of locations, our model constrains demand points such that each of them are within x minutes from a supply point. The resultant clusters can then be treated as separate distribution systems which operate autonomously
Assignment of members to teams. To facilitate group work, teams should not be composed of members with vastly different attributes. Our model hence constrains members in the same team to be within a specified proximity threshold
Constrained vehicle routing
Routing vehicles should be multifaceted and include practical constraints like forbidden zones, time-window, breaks, etc. Our software is able to use Open Street Map, and present result in vehicle view for quick analysis
Minimum vertex cover
Minimise fleet size required to cover a complex network. A simple example is to cover all hallways (edges) connecting all rooms (nodes) with security cameras. For complex problems such as analysis of travel connectivity of a geographic region, our software is able to use Open Street Map to create the required network from travel time/distance thresholds
Sound spectrogram
Data analytics
Abnormal sound detection
Use one-class support vector machine which requires only normal sound for training
Fair regression/classification modelling
Use representative sampling and kernel support vector machine to reduce class imbalance and bias
Explainable modelling
Build quadratic expansion models using quantile regression and support vector machine. Models are explainable, more accurate and require only hundreds of data points
Visualise neural network mapping
Global optimisation hybrid deep learning
Anomaly detection and classification of images
Use kernel support vector machine for better accuracy. In the scenario of detecting diseased plants, from past studies, about a few hunderd images of each class is adequate
Image regression
Build regression model from images for applications such as estimating food quantity, age from X-ray, etc. For X-ray age estimation, factors such as gender, weight, etc. are important and should included. Since gender is just a single field, merging with image which is high dimensional requires careful treatment. For high-dimension low-sample size data, support vector machine is known to suffer from data piling at the boundary, which can diminish generalisability. This means that new approaches are required