This document discusses how artificial intelligence (AI) is revolutionizing industry and manufacturing through cognitive technologies. It provides a brief history of AI development and outlines several current industrial applications of AI, including predictive maintenance and quality inspection using computer vision, acoustic analytics, and edge processing. Both the opportunities and challenges of adopting AI in industry are considered. The document advocates an approach of first gathering and visualizing manufacturing data, then infusing digitization efforts with cognitive analytics to gain insights and continuously advance systems.
6. 1950
Turing test:
measures machine
intelligence
1957-1965
First attempts to
simulate human
problem solving
(General Problem
Solver)
Ende 1960er
First chatbot
1988
German Research
Center for Artificial
Intelligence (DFKI) is
established
Ab 1997
Annual Robocup
1956
The term artificial
intelligence is
introduced by John
McCarthy
First functioning AI
program
1965-1975
Little progress is made,
cutbacks are made to
AI financing
1975-1985
Public awareness of AI
research is created
through expert system
technologies (e.g.,
MYCIN)
1997
AI chess computer
becomes world chess
champion
2011
IBM develops Watson
2016
Google develops
AlphaGo
1936
Turing machine: first
machine to simulate
any computer
algorithm
Further Internet 4.0 and
Internet of Things
innovations
8. We’re entering the fourth revolution of industry and it is fully
differentiated from any that came before it
Line
Production
Electrification &
Automation
Miniaturization &
Global Scale
Cognitive
Manufacturing
Complexity
Era
1783
1870
1960
2020
Water, steam, and
conveyors;
modern materials
handling
Assembly systems:
Lighting,
electricity and
assembly lines
Embedded
systems:
Semiconductors,
computers,
information
technologies and
increase in trade
Cyber-physical
systems: sensors,
big data,
predictive
analytics,
cognitive
computing,
robotics, 3D
printing
9. Artificial Intelligence
• Logical thinking
• Making decisions in case of uncertainty
• Learning
• Planing
• Communication in natural language
• Understanding images
• All these abilities to achieve a goal
• Goal:“Machines that behave as if they
have intelligence.“ John McCarthey (1955)
10. AI current status
10
• The systems implemented today are
a form of narrow AI.
• They can only do just one thing really
good or better than humans.
• … like recognizing objects, natural
language understanding.
11. Learns from
experience
Uses what is
learned to draw
conclusions
Identifies
images
Solves difficult
problems
Understands
different
languages
Creates new
perspectives
AI SYSTEMS
12. Watson IoT / Interactive hospital rooms
Consumer AI vs Enterprise AI
15. A human instructor can
guide the robot as it
learns the
corresponding
movements.
Trained robots can
collaborate with the
human worker and
hand him/her objects.
They constantly
improve their
algorithm.
As soon as the robot
has finished learning, it
can repeat the new
movements
independently and can
automatically handle
slight changes in its
environment.
Various sensors and
cameras can provide
information to
computer-vision
algorithms that ensure
safe collaboration
between worker and
robot.
19. Collecting parametric and event data from machines,
SCADA, PLC, MES
Production- or industry-specific analytical models,
prediction of disturbances, energy consumption
optimization
Suggested solutions, optimal time for preventive
measures, optimal settings
Monitor
Predict
Optimize
Improve production processes and equipment
20. Increased Production
AI-powered root cause
analysis can speed up the
decision-making process
on improvement
measures.
AnAI-based analysis of
process information can
predict locations of
production disturbances.
Production tool data can
be linked and supplied to
theAI machine so that
optimized process
conditions can be
automatically determined.
24. Opportunities
faster decision making
better forecasting
increased efficiency
eliminate human error
help humans perform better
reduce costs/labor force
25. Challenges
resistance and cultural change
poses threat to labor intensive and
management positions
lacks empathy
lacks moral compass
increased competition
rapid technological development
training efforts
data quality
26. Infuse with
cognitiveAdvance to
analytics &
digitization
Visualize the
patterns
• Use the platform to
visualize your data
in meaningful
dashboards
• Start to see current
status and patterns
• Instrument your
equipment/assets to
collect data
• Gather already
existing data
• Gain insights from the
data
• produce analytical
models to predict
equipment failures
and provide
recommendations
• Streamline business
processes
• Refine models with
cognitive machine
learning
• Use speech, video,
image to diagnose
complex problems
• Utilize other cognitive
functions to improve
engagements
Gather the
data
Approach
Notas do Editor
----
Tabulating Systems...1880...Counting polulations in US...Hollerith
Programmable: Here System 360..first large scale c0mputing power....around 1960...2nd most expensive project in the 60s....after Apollo... 5 system 360 needed for Apollo mission....Storage equals 2 mp3 music files.....
Cognitive Era – Understand/Reason & Learn...future of IoT is cognitve
Why cognitive: Explosion of data..strucutre (ERPO and unstrutured social, image – meanwhile Brontobyte...Billion times every grain of sand on earth..data volume in 10 years....you cannot deal with that in a programmable manner...no enough code and manpower to do so
Cogitare – lat. denken
Understand, reason and learn
Everybody currently works on become digitail and industry 4.0. Fast ones will be winners...but what is if all is digital...we belive cognitive makes the difference
It took 1000 years to get 500 million people to Christianty....internet 5 years ..it took 35 days to get angry birds to 500 million people...
Why cognitive
----
Noice, pictures, video,
Brontobyte = 1 billion times every corn of sand on the planet...in 10 years amount of data we produce per year
Why cognitive: Explosion of data..strucutre (ERPO and unstrutured social, image – meanwhile Brontobyte...Billion times every grain of sand on earth..data volume in 10 years....you cannot deal with that in a programmable manner...no enough code and manpower to do so
Cognitive systems are fundamentally different from what we have today
Everybody currently works on become digitail and insustry 4.0. Fast ones will be winners...but what is if all is digital...we belive cognitive makes the difference
Unstructured data — “dark data” — accounts for 80% of all data generated today.
Es gibt immer leistungsstärkere Systeme, um die Informationen sinnvoll zu verarbeiten
Auch mit der künstlichen Intelligenz haben wir uns schon seit sehr vielen Jahrten beschäftigt
Den Begriff künstliche Intelligenz und entsprechende Lösungen gibt es schon sehr lange.
Aber erst in der jüngsten Vergangenheit wurde der Einsatz immer verbreiteter.
Das hat u.a. mit den fallenden Kosten für die benötigten Komponenten zu tun.
Früher gab es bereits Extertensystem, regelbasiert, sehr aufwändig zu programmieren.
Um die Programmierung zu vermeiden, werden heute Systeme eingesetzt, die die Erkenntnisse aus den Daten selbstständig gewinnen.
Dazu sind natürlich viele Daten nötig.
We’re entering the fourth revolution of industry, the Cognitive Manufacturing era, and it is fully differentiated from any that came before itDigital transformation of production processes create new opportunities to achieve levels of productivity and specialization not previously possible
Benefits include:
Higher automation and productivity of manufacturing process
Increased quality and competitiveness across the value chain
Ability to create new markets by establishing new services and business models
Logisches Denken Treffen von Entscheidungen bei Unsicherheit Lernen Pl…
Und damit kann künstliche Intelligenz in der Praxis erst sinnvoll eingesetzt warden
Das sind einige Tätigkeiten, die KI-Systeme durchführen können
Zu beachten sind dabei Unterschiede zwischen KI für Consumer und Unternehmen
---
Ownership of Data
Industry process skills
Security & Trust
Ein Anwendungsfall ist zB Industrieroboter.
In diesem Fall lernen die Greifer im laufenden Betrieb und werden dabei – mit Unterstützung von Menschen – schlauer.
Generelll kann man sagen, dass die Systeme durch Verwendung der vielen Daten schlauer und die Programmierungsaufwände reduziert werden.
Mitsubishi Robot
MELFA RV-7FLM
1. IBM Visual Insights
We are using our Visual Insights product to perform automated quality inspection on the BMW spare parts. The inspection is done based on the pictures that the camera at the tip of the robot is taking.
2. IBM Predictive Maintenance
Using our Predictive Maintenance solution, we are detecting subtle anomalies in the robot's operating behavior which may be early indicators for future breakdowns. For that purpose, we built up a model using high resolution servo current data representing normal behavior. Using our custom MQTT Gateway we are now forwarding actual operational data to the model for scoring. Subtle changes in the servo current patterns are then picked up and raised as alerts by Predictive Maintenance.
3. IBM Acoustic Insights
We are using the Acoustic Insights solution to detect anomalies in the robot's operating noise. For that purpose, we are attaching a contact microphone to the robot base that's picking up its sound patterns. Using Acoustic Insights we already trained a model based on "normal" noise during operation. We are then applying the model to pick up abnormal sound patterns that we provoke on stage.
4. IBM Equipment Advisor
Equipment Advisor is bringing the different pieces together. It provides a single entry point to access operational, historical and supporting information about the robot and even interact directly with it. In our case, the single entry point is a chatbot that allows access to these data points and functionality. In the background, Equipment Advisor is correlating real-time operational data with unstructured data from operating manuals and presents it in a curated and targeted way.
Erkennen von Qualitätsproblemen in der Produktion oder im Betrieb über unstrukturierte Daten (Bilder, Töne, Vibrationen) und analytische Verfahren (Frühwarnsysteme)
Goal
Automated diagnosis of dishwasher spray arms in final test via acoustic analytics in the factory environment
Challenges
Unobservable in normal working state with door closed
Need abnormal operation to test with door open in working state
Time-consuming and complicated
Value to client
Enable new test function for normal working status with (door closed) that is unobservable in current status
Ensure the high quality product with high accuracy of final test by overcoming the limitation of experience and physical status of test engineers
Explore the new tecnologies that could be widely appied to other such as microwave oven, washing machine, range hoods, etc. and spanning the whole product lifecycle from production to after-sale service
Schaffler winfmill bearing
SMCF bearing at train
Verbessern von Produktionsprozessen anhand wesentlicher Kennzahlen (OEE) mit Hilfe von Industriespezifischen Modellen für diskrete Fertigung und Prozessindustrie.
Ein Beispiel für eine modulare Fabrik ist die SmartfactoryKL.
Dort werden von den Modulen der beteiligten Unternehmen Visitenkartenhalter hergestellt.
Ist bei der Hannover-Messe zu sehen und kann auch besucht werden.
Roboter transportiert Werkstücke.
Users of the Cognitive Factory get the status of the production process either by using the GUI or by chatting in English or German.
Again, the PSB is used to integrate productiomn systems the cloud-based solution.
Instructions and more Information can be found here (i).
Bei Ihrem Einsatz ergeben sich eine Reihe von Vorteilen
Allerdings gobt es auch eine ganze Reihe von Herausforderungen