Engineers and scientists are finding new and creative applications for data analytics technologies. They use machine learning, big data, and optimization to make critical decisions in medical diagnosis, stock trading, energy load forecasting, and more. In MATLAB® they can acquire and preprocess data from a variety of sources, build predictive models, compare machine learning algorithms, and integrate algorithms into production systems.
Training neural networks to predict maintenance requirements for gas and oil extraction equipment
Baker Hughes trucks are equipped with positive displacement pumps that inject a mixture of water and sand deep into drilled wells. With pumps accounting for about $100,000 of the $1.5 million total cost of the truck, Baker Hughes needed to determine when a pump was about to fail. They processed and analyzed up to a terabyte of data collected at 50,000 samples per second from sensors installed on 10 trucks operating in the field, and trained a neural network to use sensor data to predict pump failures. The software is expected to reduce maintenance costs by 30–40%—or more than $10 million.
Optimizing HVAC energy usage in large buildings
Heating, ventilation, and air-conditioning (HVAC) systems in large-scale commercial buildings are often inefficient because they do not take into account changing weather patterns, variable energy costs, or the building’s thermal properties. BuildingIQ’s cloud-based software platform uses advanced algorithms to continuously process gigabytes of information from power meters, thermometers, and HVAC pressure sensors. Machine learning is used to segment data and determine the relative contributions of gas, electric, steam, and solar power to heating and cooling processes. Optimization is used to determine the best schedule for heating and cooling each building throughout the day. The BuildingIQ platform reduces HVAC energy consumption in large-scale commercial buildings by 10–25% during normal operation.
Czech Academy of Sciences
Developing algorithms to reduce false alarms in ICUs
False alarms from electrocardiographs and other patient monitoring devices are a serious problem in intensive care units (ICUs). Noise from false alarms disturbs patients’ sleep, and frequent false alarms desensitize clinical staff to genuine warnings. Competitors in the PhysioNet/Computing in Cardiology Challenge were tasked with developing algorithms that could distinguish between true and false alarms in signals recorded by ICU monitoring devices. Czech Academy of Sciences researchers won first place in the real-time category of the challenge with MATLAB algorithms that can detect QRS complexes (deflections of the heartbeat trace from its baseline in an ECG), distinguish between normal and ventricular heartbeats, and filter out false QRS complexes caused by cardiac pacemaker stimuli. The algorithms produced a true positive rate (TPR) and true negative rate (TNR) of 92% and 88%, respectively.
Analyzing big data for online freight logistics
Freightos developed an online freight routing and pricing system that uses Google® BigQuery to manage and store multiple databases for thousands of freight contracts, millions of freight quotes, and a wide array of other shipping data. Engineers export results from queries performed on BigQuery to the cloud, where they are downloaded and analyzed in MATLAB. One analysis evaluated 120,000 rows of freight quotes, each with more than 30 columns, to identify fluctuations in freight pricing based on different sales teams, companies, and shipping modes.
Statistics-based health monitoring and predictive maintenance for manufacturing processes
Mondi Gronau’s plastic production plant delivers about 18 million tons of plastic and thin film products annually. Machine failures that result in downtime and wasted raw materials cost millions of euros each month. To minimize these costs and maximize plant efficiency, Mondi developed a health monitoring and predictive maintenance application that uses bagged decision trees and other machine learning algorithms to identify potential issues, enabling workers to take corrective action before a machine breaks down. A standalone executable version of the application is now used in production at the plant.
Developing virtual metrology technology for semiconductor manufacturing
Photolithography is the process used to create the patterned layers of a silicon microchip. ASML’s TWINSCAN photolithography system uses overlay control to ensure precise alignment of the layers. So many overlay marks are required for proper overlay model correction that it is not feasible to measure every wafer coming out of a TWINSCAN system. ASML developed virtual overlay metrology software that applies machine learning techniques to come up with a predicted estimate of overlay metrology for every wafer, using alignment metrology data. The network identified systematic and random overlay errors that might otherwise have gone undetected.
Aberdeen Asset Management
Implementing portfolio allocation models in the cloud
Aberdeen Asset Management bases many of its trade decisions and multi-asset class mandates on portfolio models generated with advanced machine learning algorithms. Analysts trained the models using factors such as monetary policy, corporate profits, interest rates, and implied volatilities. They backtested the trained models on more than 15 years of historical data using MATLAB Distributed Computing Server™ in the Microsoft® Azure cloud. After refining the techniques and incorporating them into their asset allocation algorithms, they now run the algorithms with large financial data sets on a distributed computing cluster.