Workplace-based Learning in Industry 4.0 -- Multi-perspective approaches and solutions for the shop floor

Workplace-based Learning

in Industry 4.0 Multi-perspective approaches and solutions

for the shop floor

Carsten Ullrich

Associate Head

Educational Technology Lab (EdTec),

German Research Center for Artificial Intelligence (DFKI GmbH)

Deutsches Forschungszentrum für Künstliche Intelligenz

German Research Center for Artificial Intelligence

• One of the largest research institutes in the field of innovative software technology based on AI methods

• Focusing on complete cycle of innovation - from world-class basic research through prototypes to product functions and commercialization.

• Research and development projects are conducted in 10 research departments, 10 competence centers and 5 living labs

• Educational Technology Lab– Support of education and qualification processes

through innovative software technologies

– Research, development and consulting

– Focus on technologies that intelligently adapt and adjust learning environments and learning materials to individual learners

– http://edtec.dfki.de/

Carsten Ullrich, Workplace-based Learning in Industry 4.0

Towards Industry 4.0

tEnd of18th Century

Start of 20th Century

First Mechanical

Loom1784

1. Industrial Revolutionthrough introduction of

mechanical production

facilities powered by

water and steam

2. Industrial Revolutionthrough introduction of mass

production based on the division

of labor powered by

electrical energy

Start of 70ies

4. Industrial Revolutionbased on Cyber-Physical

Production Systems

today

010001101001010100100101010010010101

Industry 1.0

Industry 2.0

Industry 3.0

Industry 4.0

De

gre

e o

f C

om

ple

xit

y

3. Industrial Revolution electronics and IT and heavy-

duty industrial robots for a

further automation

of production

Wahlster, 2012Carsten Ullrich, Workplace-based Learning in Industry 4.0

The Workplace is

Transforming

• Challenges for Europe's manufacturing industry:– Accelerating innovation

– Shorter product cycles

– Ever increasing number of product variants

– Smaller batch sizes (batch size 1)

– … while keeping/increasing level of competitiveness

– … with fewer and fewer employees


Human Operators at

Tomorrow’s Workplace

• Despite the increasing automation, human operators have place on shop floor with changed roles

• Contradictory predictions:

"Optimistic view"– Job losses compensated by new jobs

– “Better” work, increased qualifications

– Higher autonomy and self-organization

"Pessimistic" view– Major job losses

– Polarization as middle layer disappears

– Advanced control


(Source: Hirsch-Kreinsen, 2017)

What do we want?

What does our

technology enable?

Sociotechnical Perspective

• Technological innovation cannot be considered in isolation, but requires an integrated approach drawing from technical, organizational and human aspects.


Technology

OrganizationHuman

Assistance- and Knowledge-Services

for Smart Production

• Challenges– Industry 4.0 increases complexity on the shop floor

– Employee under constant pressure• to solve problems occurring on the shop floor as fast as possible,

• to improve work-related knowledge, skills, and capabilities

• Chances– Industry 4.0: sensors, actors, data

• Opportunity to build tools that– adapt themselves intelligently to the knowledge level and tasks of the

human operators

– integrate and connect the knowledge sources available in the company

– generate useful recommendations of actions

– enable recording of work processes and applied knowledge

– support the migration towards smart manufacturing


APPsist Consortium

Ap

plicati

on

&

Valid

ati

on

Researc

h &

Dev

elo

pm

en

tC

on

su

ltin

g

*Subcontracts

*

Duration 1.1.2014-31.12.2016




Partly automated assembly

line

Support for maintenance

5-axis drill

Support for machine usage

Pilot Scenarios

Partner

Pilot Area

Pilot Scenario

Production line

Support for failure detection


3 manual assembly

stations

Main host computerMonitoring and analysis

SPSControlling the machines

Coarse control and

monitoring granularity

System detects status and

faults

Classification on level of

stations, not components

Activities

Preventive maintenance

Resolving disabled states

and faults

Manual assembly

Goal

Increase scope of actions of

workers

Increase workers’

understanding of process,

product, manufacturing

Automated processes

Machine user

Machine operator

(plus)

Machine operator

Com

pete

nce

Pilot Study: Festo


Pilot study Festo: Refill Loctite


Characteristics of Support


MENSCH-

MASCHINE-

INTERAKTION

• Knowledge discovery:

Recommend relevant

information

• Notification: Inform

employee that relevant

information is available

For the employee:

• Support work

procedures

• Widen range of actions

• Gain experience

• Gain insights

• Make work meaningful

Company:

• Increase flexibility

• Increase productivity

Translation into concrete

requirements: joint work

with work council and I4

experts from union

Control lies in hands of employee

APPsist‘s Assistance- and

Knowledge Services

• APPsist: First general applicable service-oriented architecture, with company specific

specializations

– Machinery, job profiles, learning materials, documents, ...

Smart Services: Use of existing infrastructure to implement new functionalities

• User-centered: Focus on support, qualification, further training of the employee

• User-adaptive, context-based support through formalized expert knowledge


Databases

Machinery Employees

DevicesAR

Smart Services

Basic Services

Assistance in

carrying out activities

• Objective: Perform work activities as efficient and effective as possible– Achieve production targets (OEE, Overall Equipment Effectiveness)

• Contextual recommendations by displaying– Relevant work activities

– Relevant information (circuit diagrams, construction blueprints, manuals, ...)

• Assistance during activity– Display of the individual steps of an action (step-by-step instructions)

– Augmented Reality: superimposition of information in the field of vision

– Adaptation using sensor data


Supporting Learning

• Performing a work procedures does not automatically

lead to learning

• Goal: Support targeted knowledge acquisition

– Display relevant work procedures

– Display of relevant content and information (learning materials,

manuals, ...)

• product

• production

• process

• Taking into account

– Performed work procedures

– Development goals






Artificial Intelligence in Education

• Intelligent Tutoring Systems and Adaptive Learning Environments provide adaptive and contextualized support of learners

• Significant body of research on adaptive support in university and highly structured domains such as mathematics, physics and computer science

• Methods– Knowledge-based systems:

Modelling human experts

– Statistical approaches


Domain Model

Learner Model

Pedagogical Model

APPsist Ontology

• Describes relevant concepts for and their relationships

• User

• Content

• Manufacturing

• Representation in OWL (Semantic Web standard)

• Used for communication between services and for reasoning by intelligent services


User Model

• Connection to domain-model concepts• Concepts from domain-model are enriched with user specific

values– Number of executions (for process-steps)

– Number of views (for contents/documents)

– Number of usages (manufacturing/production objects)

• Relevant user properties• Workplace-groups

• Permissions

• “State“: main activity (KPI), secondary activities

• Development goals

• Mastered measures


Adaptivity in Smart

Manufacturing

• Main activity: Fulfill Key Performance Indicators (KPI) Assistance: Depending on the contexta) Reacting to the current situation on the shop floor, e.g.,

Loctite is empty

• Secondary activity: Time for Learning Learning: Depending on the employeeb) Reacting to recently occurring events (e.g., a large number

of correctly or incorrectly performed measures)

c) Long-term development goals (e.g., working towards a new job position)


If employee is in state “main work activity” and asks for assistance, then

select work procedures relevant for current station und machine state:

1. WU = workplace unit to which employee is assigned to.

Determined through request to user-model-service.

2. S = sort(stations ∪ installation) of AG. Determined by querying

domain model: There, each workplace unit is assigned to work with

specific installations. An installation consists of stations. Sort the

stations according to priority of each station.

3. MS = machine state of S, sorted according to priority of machine

state. Determined through request to machine-information-service.

4. P = Procedures for MS. Determined through query of domain

model: Procedures are applicable to machine states.

5. P_a = those procedures of M the employee is authorized to

perform (with or without assistance). Determined through request

to user model.

Result: P_a

Select Measures, Main Activity

Examples

1. WU = (Production

of standard

cylinders)

2. I =

(DNC_DNCB_DSB

C, …) . Stations =

(S10, S20, …).

Pri(DNC)=8

3. MS = (LociteEmpty,

GreaseFew, …)

4. P = (ChangeLoctite,

ChangeGrease, …)

5. P_a =

(ChangeLoctite)


If the employee is in state secondary activity (“time for learning”) and asks for

procedures, then select procedures relevant to development goals (content C_A,

and/or position PO, and/or production items PI_A).

1. PO = agreed future position of employee. Determined by query to user model.

2. P = relevant work procedures for PO. Determined through query to domain

model: Each position has tasks, and work procedures perform tasks.

3. P_U = P without mastered procedures. Determined through query to user model

(which keeps track of mastered procedures).

Result = P_U.

Select Measures, Secondary Activity


If the employee is in state “main work activity” and asks for information, then select

content relevant for the stations assigned to and their machine states:

1. WU = workplace unit to which employee is assigned to; P = position of

employee. Determined through request to user-model-service.

2. S, MS = Machine states and stations/installations relevant for WU (see

previous rule)

3. I = Content about S∪MS for target-group = P or without target-group.

Determined by querying domain model, which contains metadata that relates

content to domain model entities and specifies its target-groups, if any.

Result = Content I.

For instance: operation manuals, circuit diagrams, and other content that provides

information about the current situation enabling the employee to overcome

occurring problems.

Select Content, Main Activity


If employee is in state secondary activity (“time for learning”) and asks for content, then select

content relevant to current work history (machines and procedures worked with). Development

goals: content C_A, and/or position PO, and/or production items PI_A.

1. PI = production items with which employee has worked with in the last four weeks, P_S the

procedures that she performed successfully and P_N those not performed successfully.

This information is stored in the learner-record-service.

2. C_P_N = content about P_N and production items used by P_N, with already seen content

sorted to the back (this information is stored in the learner-record-service).

3. C_P_S = content about P_S or about production items used by P_S or about PI.

4. C_P = Content that covers one/several of the following: position PO, tasks of PO, or

production entities PI_A.

5. C_PI_PO = Content that describes production entities relevant for PO.

6. C_P_PO = Content that describes production entities used for performing procedures

relevant for PO.

7. C_T = C_P_S ∪ C_P ∪ C_PI_PO ∪ C_P_PO, with already seen content sorted to the

back.

Result: Content C_P_N + C_A + C_T, with duplicates removed.

Select Content, Secondary Activity


?? State of the Art ??


Are rule-based

systems state of the

art?

AIED in Industrial Production

• AIED in Mathematics / Physics: More than 30 years of research

– Principles well understood

– Proven architectures

• Learning at the industrial workplace:

– Multitude of single systems, no common basis

• APPsist:

– First general ontology (domain description) with focus on learning in

production environments

– First general rules to support the employees

• Rule-based systems have proven themselves, well-understood for

which problems they are suitable

• Statistical approaches require data…


Digital Education Space

• Learning systems can easily capture actions of the learner

• Data is simple but usable (click data, performance)

• Learning Analytics: Real-time recognition of learning progress, motivation, correlations between navigation behavior and learning success

Update of learner model

Feedback to learners and teachers through a pedagogical model


Analogue Education and

Work-Spaces

Analogue Spaces out of reach

for learning systems

No learner modelling and

adaptive reactions possible


Stepping from the

Analogue into the Digital

• APPsist:

– First steps towards the

use of data signals from

"analogue" world (sensor

data of the production

plants),

– their interpretation

regarding the actions of

the employees,

– and their usage for

automated support


Internet of Things for the

Digitization of Existing Spaces

• Increasing penetration of environments with sensors / actors

– Smart Factory

– Smart City

– Smart Home

– Smart Energy

• Usage of Smart Data also for user-centered support

• Coupling between work- and education spaces


Coupling between Work- and

Education Spaces

• In the education space:

learning adapted to

activity and goals

• In the workspace: during

the execution of activities

references to relevant

training materials

• Authoring support (EdTec

Project DigiLernPro)

• Data collection

statistical methods!

Smart Training Services


Required:

• privacy and data protection

• design principles: enable good work and good learning

Sociotechnical Perspective!

Thank you

Carsten Ullrich

[email protected]