How Prescriptive Analytics Empowers IoT to Reduce Operational Risk, End Breakdowns, and Increase Net Process Output and Profitability Wednesday, October 7, 2015 Presented by Mike Brooks President & COO, Mtell
Learn Adapt Inoculate
Session Description The presentation confirms how machine learning, when industrialized and applied right, is a simple way to augment and even replace costly and ineffective old practices. The session shows how software agents do the heavy-lifting, do the monitoring, automatic analysis, and accurately predict and alert so that you do not have to. With little human help they learn, train, and adapt. And, triggered by the earliest alerting available, the prescriptive action completely changes outcomes to avoid issues and/or alleviate the consequences. The result is more stable operating plants and consistent high net output at lower cost. 3
About the Speaker Mike Brooks President & COO Mtell 5 start-ups >>> 25 + years bus. dev., prod marketing & mgmnt., process ops., planning and scheduling, control & IT systems Chevron Technology Ventures invested in technology companies board seat positions at tech companies Sr. control engineer & Industrial IT infrastructure, historians, dashboards, data warehousing Led 4 IT start-ups invented manufacturing solution, & founded INDX Software to commercialize product Steered Wonderware Products & Technology vision Control, Historian, and UI products VP Marketing, and user requirements at OpenO&M
Predictive Analytics & San Diego a global capital for predictive technologies
History 1986, HNC target detection for DoD & quickly adapted for credit card fraud detection Spawned dozens of companies SAIC & L3 Communications help identify terrorists Kuity Corporation search Medicare records for fraud Detectent mine utility records for illegal power grid connections Mtell watch real-time signals for early signs of machine failure 80+ predictive analytics companies
Machine Learning It s not about machines But computers emulating human characteristics
Machine Learning a scientific / engineering discipline that deals with the construction and study of algorithms that can learn from data. Such algorithms operate by building a model based on inputs and using that to make predictions or decisions, rather than following only explicitly programmed instructions.
Machine Learning support vector machines Geoffrey Hinton University of Toronto deep learning networks Now dominates all fields of predictive analytics
Critical incidents don t just happen They happen because we keep doing things the same way
What s Being Challenged Industrial maintenance best practices Run to failure Calendar Usage Condition RCM
Current State of Maintenance 85% equipment fails in spite of calendar maintenance Boeing 63% scheduled maintenance is unnecessary Emerson
What if Change Looked like This Old maintenance approach New maintenance approach
The Industrial Revolution is over It s time for the Industrial Resolution
The Industrial Resolution Breakdowns Environmental Catastrophes Casualties
Prescriptive Maintenance no different than a visit to the doctor
Prescriptive Maintenance Symptoms analysis Diagnosis Consideration of treatments Prescription for action
Algorithms are only 10% of the solution Learn Adapt Inoculate
Change the Strategy Instead of the basket of data in the sky with 6 experts stirring Get the software to do the work so you don t have to
Use Agents to Do the Work for You To embed learning and adaptive big data, predictive, and prescriptive analytics Conscious Monitoring Agents Where each one is better than an expert person interpreting every failure signal, every minute of the day forever and tell you what will happen, when, and how to avoid it
Replace the Best of Yesterday s Technology Eliminate the inaccuracies and false positives No More False Alarms Where every message is a truthful, early, accurate forecast you can trust delivering good prescriptive advice that you can rely upon and take affirmative action
Eliminate Models, Agents, Rules Learn from actual sensor data and retain the knowledge forever Transfer Protection to Others Where every behavioral aspect you learn transfers; applying the same breakdown, safety, and environmental protection to all similar equipment
Making Machines Smart Learn Adapt Inoculate Agents learn from their surroundings Agents watch and warn of new danger Agents share and increase protection
The The Internet Internet of of Things (IIoT) The Collection of: One element of the Industrial Internet of Things Sensors Sensor Networks Smart Machines Computer Power Analytics People Solving problems that were previously unsolvable
New Solutions Ask New Questions OLD Is this normal NEW What precise pattern led to this Do I recognize this pattern examines new conditions against estimates of normal makes estimates from models, equations, and conditional rules anomaly detection is imprecise and always needs human interpretation great skill to build, error prone, fragile, and difficult to maintain examines new conditions against precisely known patterns patterns come from real, hard, monitored, collected sensor data anomalies detected precisely new data do not match known patterns predictive/prescriptive agents are made in minutes by end users
sensor time-series Training Agents normal failure condition precise time-to-failure inspection work orders earliest signature failure from EAM
Types of Assets Industrialized machine learning based prescriptive analytics works on more than rotating equipment Pumps & Compressors Motors & Drives Electrical Transformers Vehicles Process Equipment Centrifugal Reciprocating Rotary Centrifugal Reciprocating Rotary Plus Other Static Equipment Heavy Haul Trucks Locomotives Chillers Ht. Exchangers Boilers Best Practices for Oil & Gas 2015 28
Smart Machines tell you WHEN & HOW an asset will fail < Mtell Smart Machines > Anybody can make that claim can they really do it
Industry Results Transportation Water / Wastewater Refining Pharma Upstream Mining Pulp & Paper Chemicals
Refining Case Study Decades-long issues with compressors failing in spite of $millions on vibration systems & RCM
Cast a Wider Net Cause Symptoms Consequences wet gas cooler dry gas to pipeline compressor condensate separator target the cause not the symptoms monitoring detects symptoms but symptoms of what?
Refining Case Study 1 Conscious Monitoring Agents cast a wider net around equipment to cover process & machine issues Root cause detected = Liquid carry-over Chief rotating engineer: You can do that? I cannot apply enough people to do that level of monitoring & analysis.
Refining Case Study 2 Third-stage valve issues detected Alert to Cause = 8 weeks: 7 weeks before vibration suggests damage Savings: $X MM Repairs and Loss of Product
Transportation Case Study Perennial failures with locomotives cost $millions in repairs & lost revenue
Transportation Case Study Conscious Monitoring Agents scan lube oil data Engine passes low pressure test but Mtell sees failure signature management baulks High pressure test confirms
Transportation Case Study Lube oil chemistry detects degradation and failure Previous science project detected very little Savings: $10MM during trial projected $200MM over the next 2 years
Smart Machine Results Machines stop breaking down Machines last longer Net output increases dramatically Maintenance costs decrease dramatically
Real Prescriptive Maintenance Is the solution that: Does not require you to know more than already know Uses agents to do what experts cannot do Converts anomaly alerts into more accurate failure alerts Can tell you WHEN and HOW an asset will fail Inoculates similar assets with the same safety and breakdown protection
Customers Report 10x to 100x+ in 12 months
Every Minute that Goes by is costing you money and exposing you to unnecessary RISK
thank you for your time MBrooks@Mtell.com