Machine Learning and Artificial Intelligence Advancements for - - PowerPoint PPT Presentation

SEPTEMBER 5 - 7, 2018

Machine Learning and Artificial Intelligence

Advancements for Electrical Inspection

SEPTEMBER 5 - 7, 2018

Publicized Milestones

• January 2011 Watson beats long standing

Jeopardy Champion

• Amazon’s Jeff Bezos – “Big trends are not

that hard to spot. We are in the middle of one right now. Machine Learning and Artificial Intelligence.”

SEPTEMBER 5 - 7, 2018

Facial Recognition

SEPTEMBER 5 - 7, 2018

AI is Broadest of Terms

SEPTEMBER 5 - 7, 2018

Artificial Intelligence

• Mimic Human Intelligence

– Using Logic, If-Then Rules, Decision Trees

• ML statistical techniques that enable machines to

improve a task with experience

• Big data is large volumes of data used for

computational analysis, to reveal patterns or trends

SEPTEMBER 5 - 7, 2018

Neural Network

• A computer system designed to work by

classifying information similar to the same way a human brain functions

• Taught to recognize images

– classify the images, including the components or sub- elements – Probability statement, decisions or predictions with a degree of certainty or statistical probability

SEPTEMBER 5 - 7, 2018

Glossary of Terms

• Artificial Intelligence (AI) – A.I. is the broadest of term, applying to any technique that enables computers to mimic human intelligence, using logic,

if-then rules, decision trees and machine learning.

• Machine Learning (ML) – The subset of A.I. that includes statistical techniques that enable machines to improve at task with experience.
• Big Data – Large volumes of data sets that are used for computational analysis, to reveal patterns or trends.
• Deep Learning – The subset of machine learning composed of algorithms that permit software to train itself to perform tasks, like speech and image

recognition, by exposing multilayered neural networks to vast amounts of data.

• Neural Networks – Software constructions modeled after the way adaptable networks of neurons in the brain are understood to work, rather than

through rigid instructions predetermined by humans.

• Bayesian Networks – A probabilistic graphical model or type of statistical model that represents a set of variables and their conditional
• dependencies. For example, a Bayesian network could represent the probabilistic relationships between electric outages and a individual electric grid

component anomaly. Given the component anomaly type, the Bayesian Network can be used to compute the probabilities of an outage.

• Clustering – During the supervised learning phase of inputting data for training, a subject matter expert selects or targets input features to be utilized

in the learning models. In clustering or unsupervised learning, the target features are not given in the training examples. The goal is to construct a natural classification that can be used to cluster the data. The general idea behind clustering is to partition the examples into clusters or classes. Each class predicts feature values for the examples in the class. Each clustering has a prediction error on the predictions. The best clustering is the one that minimizes the error.

SEPTEMBER 5 - 7, 2018

Big Box Thought Leaders

• IBM (Watson)
• Microsoft (Azure)
• Amazon (AWS)
• Intel
• Google
• Apple
• Facebook
• Spotify
• Uber
• Salesforce

SEPTEMBER 5 - 7, 2018

Inherent Advantages

• Speed
• Accuracy
• Repeatability
• Lack of bias

SEPTEMBER 5 - 7, 2018

Origin of Project SiMON

• Scientific American
• Century old children’s game: Simon Says

– Series of commands to eliminate players

• 70’s memory game with sounds that increase

in complexity with each successive sequence

SEPTEMBER 5 - 7, 2018

Commercial AI & ML Engines

• The software developed to query the AI

engines consist of a library of Application Programming Interface (API) models

• 500 GB per second or million books

SEPTEMBER 5 - 7, 2018

Methods of Acquisition

SEPTEMBER 5 - 7, 2018

UAS Platform

• 40 Minute Endurance
• Multi Sensor Platform
• 30 mph Wind Endurance
• Autorotate for Safety
• Less than 55 lbs.
• ~200 ft. AGL
• 15 lbs. Payload
• Made in USA

SEPTEMBER 5 - 7, 2018

1 Flight – 4 Datasets

SEPTEMBER 5 - 7, 2018

Flight Profile

• 50’ Above Structure
• 33’ Offset
• Left or Right Centerline
• Down & Back & Following
• Geofence Restrictions

SEPTEMBER 5 - 7, 2018

Thermal Imagery (IR)

SEPTEMBER 5 - 7, 2018

Virtual Side-by-Side Analysis

SEPTEMBER 5 - 7, 2018

Corona (UV)

SEPTEMBER 5 - 7, 2018

Close Range Oblique Still

SEPTEMBER 5 - 7, 2018

LiDAR 50 ppsm

SEPTEMBER 5 - 7, 2018 Collect Remotely Sensed Datasets Prioritize Corrective Action What? Where? Why? When? Generate Anomaly Report Inspection Summary Report

General Process Model

SEPTEMBER 5 - 7, 2018

Component Analysis Predictive Analytics

Visible Infrared Ultraviolet LiDAR

• Process & Procedures
• Software
• Filtering & Analyzing
• Large Volumes of Remotely

Sensed Data

Inputs & Process

SEPTEMBER 5 - 7, 2018

1000’s Images for Training

SEPTEMBER 5 - 7, 2018

Anomaly Reports

SEPTEMBER 5 - 7, 2018

Current Progress

• Supervised Phase – Current and ongoing

expected to last 2.5 to 3 years.

• Transition Phase – Incremental reliance on AI &

ML, with Subject Matter Expert verification. Current and ongoing with a 10% to 15% automated.

• Unsupervised Phase – 3 to 5 years to achieve an

85% to 95% automation with Subject Matter Expert providing Quality Assurance.

SEPTEMBER 5 - 7, 2018

1,105 Miles & 6,927 Structures

SEPTEMBER 5 - 7, 2018

Ground & Aerial Inspection

SEPTEMBER 5 - 7, 2018

Issues

SEPTEMBER 5 - 7, 2018

Issues

SEPTEMBER 5 - 7, 2018

Project Statistics

Oblique Still Photos

• 254,000+ photos
• 3.8 TB

Oblique Thermal Imagery

• 254,000+ images
• .40 TB

Vertical Nadir Imagery

• 169,000+ photos
• 2.5 TB

LiDAR

• 50 ppsm
• 11.5 TB

SEPTEMBER 5 - 7, 2018

A I Humor