AI/ML doesn’t work on raw operational data. What’s needed is standardized data. Sight Machine makes plant data ready for data science.
Artificial Intelligence & Machine Learning
Representative AI/ML Applications
Use AI/ML on the Sight Machine platform, or do it yourself on standardized data.
Use Case
Analyze product, temperature, speed, and machine setup data to identify factors that affect quality for 4 plants supplying EV vehicles with precision-engineered electrical components.
Techniques Applied
Gradient Boosted Regression
Application
Predictive Quality
Use Case
Combine warranty claim data with production data to do root cause analysis on factors affecting warp in boxboard production.
Techniques Applied
Gradient Boosted Decision Trees
Application
Predictive Quality, Energy use
Use Case
Provide predictions with high degree of accuracy up to three days in advance of potential quality problems in furnace for float glass line. Minimized energy use per unit of saleable glass by optimizing quality.
Techniques Applied
Support Vector Machines
Application
Asset and production scheduling
Use Case
Combine daily production schedules, real-time plant floor data, and highly variable raw material arrival to dynamically optimize asset utilization and minimize energy use.
Techniques Applied
Genetic algorithms
Application
Performance optimization
Use Case
Incorporating client chemistry-based modeling and statistics to create predictions of fermentation time on cheese production time. Used to maximize output and yield while maintaining quality.
Techniques Applied
Monte Carlo simulations, GLM regression
Application
Predictive Maintenance
Use Case
Perform anomaly detection on 300 robots on automated final assembly process (Body Shop) to provide predictions and root cause analysis for downtime.
Techniques Applied
Breakout Analysis: e-divisive with medians (anomaly detection)
Application
Downtime Prevention: Micro-stops in high-speed packaging
Use Case
A simple analysis of alarms was not identifying the true root causes of alarms in packaging lines. Problems had been unresolved for 20 years. Used a combination of AI techniques to trace problems to true root causes in real-time.
Techniques Applied
NLP, Agglomerative clustering,
Sequence analysis
Application
Predictive Downtime
Use Case
Analysis of 500 million data points from 600 welding robots to provide early warning of potential downtime and other welding faults.
Techniques Applied
Functional Data Analysis
Application
Downtime prevention: Microstop and alarm analysis to identify root cause of downtime
Use Case
Analysis of 10,000 alarms across automated, integrated contact lens line. Identified causation of frequent micro-stops with high level (95% +) predictive power.
Techniques Applied
NLP, Agglomerative clustering, Sequence analysis
Application
Quality: Computer vision inspection on vehicle final assembly line at a $12B/yr Detroit assembly plant.
Use Case
Continuous inspection on the final assembly line for Jeep Grand Cherokees and Dodge Durangos: 1,100 vehicles per day, inspecting 15 exterior elements, 25+ models, 11 colors, 24/7 operation, 99.9% accuracy. Integrating MES data, image analysis, ML and edge/cloud compute.
Techniques Applied
Machine learning continuously applied to large image sets analyzed with Simple CV, a company – developed open source CV framework
Manufacturing’s Opportunity is to Go Beyond Pilots and Deliver Scale, Sustainability and Impact
Pilot Purgatory
Among enterprise manufacturers, the case for AI has been made. Still, despite the many billions invested, manufacturers remain mired in “Pilot Purgatory.”
The reason for Pilot Purgatory is poorly understood. It is not that manufacturers don’t believe in AI. Data is abundant, manufacturers have moved plant data to Cloud, and good algorithms and models are widely available. Instead, the main problem is the state of manufacturing data itself. Unlike modern data in most business activities, plant data is not “IT ready” and is not suited to the application of AI.
What’s Needed to Drive AI at Scale?
What’s the breakout move? Standardized data.
What creates standardized data? Standardized models constructed through stream processing.
What else is needed by a manufacturer to manage hundreds of stream processing pipelines concurrently? Pipeline management tools.
Real Outcomes In Weeks
Getting started is easy
Curious about how we can help? Schedule a chat about your data and transformation needs.
