Industrial Machine Intelligence The Golden Braid of Data Streams, AI, and Human Expertise Drew Conway – Machine Learning in Oil and Gas, Canada
Google releases Data Science Venn AlphaGo defeats “MapReduce” paper Diagram published Lee Sedol Facebook releases “DeepFace” paper Hadoop 0.1.0 Release `04 `05 `06 `07 `08 `09 `10 `11 `12 `13 `14 `15 `16 `17 Gartner drops “big data” IBM Watson from Hype Cycle wins Jeopardy Google releases “BigTable” paper U.S. Supreme Court hears arguments based on big data Alluvium | Machine Learning in Oil & Gas, Canada
18th Century 19th Century 20th Century 21st Century First programmable logic controller, Modicon 084 First mechanical (1969) loom (1784) First steam powered conveyor belt in Chicago meatpacking (1867 ) Alluvium | Machine Learning in Oil & Gas, Canada
Industry 4.0 18th Century 19th Century 20th Century 21st Century First programmable logic controller, Modicon 084 First mechanical (1969) loom (1784) First steam powered conveyor belt in Chicago meatpacking (1867 ) Alluvium | Machine Learning in Oil & Gas, Canada
The Data Problem Data Lake EXTRACT, TRANSFORM & LOAD Heterogeneous Data Streams LEARN → Streaming asset time-series 1. Resource Intensive data BUILD PRODUCTS → Asset and operation meta-data 2. High Latency EXTRACT → Historical operation databases VALUE 3. Limited ROI 8 Alluvium | Machine Learning in Oil & Gas, Canada
The Tools Problem DISCOVERY REASONING VS. Machine intelligence products often seek to replace expert operators . But, this fails by not… 1. Leveraging operator expertise 2. Limiting cognitive focus 3. Reflecting reality of operator job 9 Alluvium | Machine Learning in Oil & Gas, Canada
Data-driven Decision Making in Industrial Operations 10 Alluvium | Machine Learning in Oil & Gas, Canada
http://alluvium.io/primer
Rapid Forensic Analysis Our assets and operation generate more data than we are PAIN Reduce Complexity capable of analyzing. CAUSE Industrials leads all sectors in connected device growth at a rate of 24% per annum. The volume of data produced by this Primer is designed to distill operational technology vastly outstrips an organization’s massive streams of raw sensor and ability to effectively leverage it. production data into usable insights for expert operators. This allows your team to rapidly move through massive amounts of data to identify where and when deviations and changes are happening in data, identify the sources of those issues, and take action. 12 Alluvium Primer | http://alluvium.io/primer
Operational Transparency Our current methods for regularly reviewing operational data PAIN See Everything are laborious and error prone. CAUSE Operational technology teams must perform regular (daily) inspection and analysis of asset data. Legacy tools still Primer can analyze an entire require considerable manual inspection of large amounts of database with a single mouse click. data under heavy time and resource constraints. This allows plant operators and managers to see the stability of their production from plant-level down to a single asset or sensor. The auto-generated reports can then be used to set the priority and agenda for an entire operational team. 13 Alluvium Primer | http://alluvium.io/primer
Early Warning We cannot identify new or unique operational disruption PAIN Always Learning conditions before they occur. CAUSE New failure conditions or disruptions patterns are very Primer’s artificial intelligence difficult to identify or predict by their very nature. In addition, it can be challenging to broadcast institutional knowledge adapts to your specific operation gleaned from these incidents even after they have occurred. and assets. It does this by learning from your operators’ input every time the tool is used. By constantly refining the AI, Primer can detect subtle or new patterns in data, which provide early warnings to your operators to fix problems before they occur. 14 Alluvium Primer | http://alluvium.io/primer
How We Do It Our proprietary Stability Score™ is a simple metric that pulls together historical or real-time data and gives operators a way to quickly see changes in plant and production systems, down to a single asset or sensor, and decide on the changes that matter. Connected Sensors Real-time Connected Assets Stability Scores Historical Data Alluvium Primer | http://alluvium.io/primer
16 Alluvium Primer | http://try.alluvium.io/primer
FCCU3 Selective Catalytic Reduction (SCR) system tripped causing NOx emissions to increase. 17 Alluvium Primer | http://try.alluvium.io/primer
8/27/2013 8:30AM 18 Alluvium Primer | http://try.alluvium.io/primer
19 Alluvium Primer | http://try.alluvium.io/primer
Flaring event begins 20 Alluvium Primer | http://try.alluvium.io/primer
Mattered here What happened here 21 Alluvium Primer | http://try.alluvium.io/primer
22 Alluvium Primer | http://try.alluvium.io/primer
Systematic thermal anomaly 23 Alluvium Primer | http://try.alluvium.io/primer
Three days before flaring event 24 Alluvium Primer | http://try.alluvium.io/primer
alluvium.io demo@alluvium.io
Recommend
More recommend