Pattern Recognition for Industrial Data
Heavy industries such as manufacturing depend on maximal machine uptime. Stalling a production line results in catastrophic damages, unplanned capex and lost revenue. Discovering and identifying patterns in data from industrial sensors could allow timely intervention to prevent failures, improve yields and increase plant productivity. However, due to the large volumes and high velocity of this data, human analysis is not feasible. Less than 1% of this data is analyzed today. Falkonry automates industrial operational gains by combining advanced pattern recognition techniques on industrial data with the knowledge of domain experts, and generates real-time alerts towards improving uptime, yield, quality, and efficiency.
We are delighted to see Falkonry announce their new partnership with us and Polaris Partners. Below we briefly describe the challenges in analyzing industrial data, and Falkonry’s approach towards automating and improving industrial operations.
Challenges In Analyzing Industrial Data
Comprehensive and automated analysis of industrial data is critical to driving operational improvements in a manufacturing plant. This is because patterns in this data gathered from plants, processes, and assets form the basis for yield, uptime, and quality improvement. However, given the current volume and velocity of this data — from hundreds of thousands to millions of sensors — human analysis is just not practical.
Further, most of this data is collected in a time series format. That is, a sequence of data or events recorded along a time axis. Due to the sheer scale of industrial operations and data, Humans are unable to monitor and correlate within a data stream, let alone multiple such streams, in order to understand patterns and derive value. Even available machine learning technologies cannot be applied to reveal patterns because of the high complexity of the task. The current alternative for plant managers is to over-provision maintenance and production, but this increases operating costs.
Falkonry recognizes patterns in time series data
Falkonry’s ML-based pattern recognition understands patterns in time series data by combining knowledge of processes, systems and context from domain experts with advanced machine intelligence and signal processing. Falkonry’s software processes industrial data to understand underlying patterns by automatically discovering, clustering, classifying and labelling every occurrence of these patterns. Furthermore, it can correlate data and patterns across different time series streams to develop deeper inferences, provide recommendations in real time, and support domain experts to improve yield and overall plant productivity. Falkonry is quickly building the most diverse database of signal characteristics to enable fast and accurate cataloging of patterns.
Falkonry’s software continually learns from domain experts
Falkonry’s software develops deeper domain expertise over time by learning from domain experts. The software generates notifications when an unknown pattern arises to allow the domain expert to investigate the pattern. The expert then reviews and labels these patterns. Falkonry’s software uses this newly enriched vocabulary to label similar patterns in all existing (and newly arriving) data streams to derive even richer correlations, and identify the earliest indication of failure, for example degradation of equipment. The result is in tighter quality control, higher industrial yield and reduced unplanned downtime.
Leaders in industrial systems, signal processing, ML and enterprise software
Falkonry’s leadership team is well-suited to address the needs of industrial customers. In addition to deep domain experience of this team, we were inspired by their single-minded focus on delivering customer success. Nikunj Mehta (CEO), Greg Olsen (VP Products) and Crick Waters (SVP Business Development) have advanced degrees relevant to building powerful ML models, algorithms and tools to process industrial time series signals. The team has direct experiences in large scale operations and analytics challenges at heavy industries, and in creating world-class enterprise software products. Together they are the experts in connecting Falkonry’s cognitive pattern matching capabilities to industrial yield improvement objectives.
Partnering with Falkonry and Polaris Partners.
Zetta is focused on companies just like Falkonry that build a competitive advantage by combining unique data with intelligent systems to continually improve the value of the data. This, in turn, leads to more product usage, more data, and so on. We have referred to this cycle as the virtuous loop in previous posts. Falkonry’s approach to pattern recognition that combines expert knowledge with advanced machine intelligence fits squarely in our investment purview of intelligent enterprise software. Falkonry’s modus operandi in continually learning from domain experts, aligns perfectly with our virtuous loop narrative, and empowers Falkonry in creating the winning solution for improving the operations of our industries.
