Projects

Below are a few of the innovative solutions we’ve developed for our clients.

Where's the Snow? Optical Mapping of Runway Contaminants

Bad weather dumps water, slush, snow, ice and sleet on runways, making it harder for airplanes to take off and land safely. Typically, runways are monitored through visual inspection and manual reporting. Both are prone to errors.

In a project expected to significantly improve runway safety, we developed a camera system, and algorithms to automatically analyze the photograph, evaluate and report on the nature and location of runway contaminants.

Optimizing the Work: Automating Maintenance Scheduling

Asset-heavy industries like steel mills and oil refineries often own more than 10,000 pieces of equipment, each with its own maintenance requirements. The challenges of maintenance scheduling can lead to suboptimal results, with some pieces of equipment over-maintained, while others are under-maintained.

Our client wanted a system that would reduce the cost of maintenance by developing an optimal schedule.

We provided a set of algorithms capable of learning from the past maintenance history on both individual pieces and specific types of equipment. By analyzing more than 20 factors, we were able to develop a model to predict the efficacy of a maintenance effort on a piece of equipment, based on various parameters including the resources used, time of day, concurrent activities and the weather. When combined with an optimization engine, the model creates an ideal schedule for maintenance.

Keeping it Cool: Machine Learning for Adaptive Thermal Management

Micro data centres, small enclosures containing all the essential features of full-sized data centers, are gaining popularity. But unlike their larger counterparts, micro data centres have low thermal inertia – meaning they are subject to sudden spikes in temperature as servers ramp up from an idle state to a fully loaded one.

Traditional cooling systems are slower to respond, resulting in temperature fluctuations that reduce the lifespan of electronics within the servers.

To address this problem, we designed a cooling system control algorithm that can respond more quickly, by using adaptive predictive control (APC) instead of the traditional PID control systems used for cooling.

The APC controller learns the thermal behaviour of the system from the past, and reads many signals besides the temperature, to be able to predict temperature spikes. It then preemptively ramps up the cooling hardware, thus minimizing temperature fluctuations within the enclosure.

Lower Costs and Happier Customers: Predictive Maintenance for ISP

Cable broadband uses radio frequency (RF) signals within a coaxial cable to move data in and out of a subscriber’s cable modem. Many factors, including cracks, loose splitters and weather can impact these signals and result in service problems. The source of faults and their location is often impossible to predict, requiring a trained technician to manually diagnose the network.

We created an algorithm that logs RF data in real time, and learns from the past to identify and localize anomalies that will lead to network degradation. This provides the ISP with crucial information that can be used for the proactive maintenance of its network.

The switch from reactive maintenance, often in response to an angry customer, to proactive maintenance, is expected to provide substantial cost savings.