Quarterly Review 031609

NSF Engineering Research Center (ERC) for
Reconfigurable Manufacturing Systems (RMS)
ERC Big Three Quarterly Review Meeting

March 16, 2009 ,
The University of Michigan, College of Engineering
Project review History

TA Proj # 1 Project Title Project Reviewed j 12/13/07 03/05/08 06/06/08 09/12/08 12/05/08 03/16/09 05/18/09
Data Analysis and Causal Identification for Gear Noise Reduction in Transmission Systems Cyclic Waveform Signal Analysis for Monitoring and Control of Powertrain Manufacturing Systems Throughput Analysis of Mfg Systems with Closed Loop MHS Computer Aided Simulation Model Verification, Testing and Optimization Hardware-in-the-Loop Simulation for Verification and Validation of Logic Control Manufacturing Network Time Synchronization Best Practices (NIST Funded) Reducing Unscheduled Downtime Through Automated Event-Based Control
Development, Application and Transfer of a Network ROI Cost Calculator (on hold after 12/07)
2 I 3
II
4 5 6 1 2 III 3 4 5
Wireless Network Analysis and Testing The Reconfigurable Factory Testbed (RFT) Cylinder Bore Inspection Pore Detection in Small Diameter Bores In line In-line Vale Seat Inspection Thread Measurement Camshaft / Crankshaft Polishing Testing Y. Koren Overview #2
Meeting Agenda
2:00 2:05 2:40 2:40 3:10 3:10 3:30 3:30 3:40 3:40 4:05 4:05 4:25 4:25 4:35 Introduction Reducing unscheduled downtime through automated event-based control Cyclic C li waveform signal analysis f monitoring and control f i l l i for it i d t l of powertrain manufacturing systems Computer aided simulation model verification, testing and optimization p Break In-line valve seat inspection / PKM introduction Thread Measurement New proposal: run-out measurement of crankshaft sprocket Discussions: Acceleration of tech implementation and transfer Third-party vendors involvement format Wrap-up and next steps Adjourn Adj Reuven Katz/Hagay B. Reuven Katz/Hongwei Z Reuven Katz James Moyne Judy Jin Sam Yang
4:35 5:00
All
5:00 5 00
NSF Engineering Research Center for Reconfigurable Manufacturing Systems University of Michigan College of Engineering
Y. Koren Overview #3
Engineering Research Center for g g y Reconfigurable Manufacturing Systems
Reducing Unscheduled Downtime Through Automated Eventbased Control

Dr. James MoyneUM Prof. Dawn Tilbury UM Jeff DobskiGlobal Engine Student Lead: David Linz Supporting Students: Garima Garg Edwin Teng Deepak Sharma Garg, Teng,
QRM, March 16th, 2009 Indicates New Result
TA2-2, Slide 1
The University of Michigan, Ann Arbor
Outline
Introduction
The idea, objectives and approach
Project History C Current F t Focus

Addressing changing conditions Interim results
Next Steps
Short term Longer term project planning Summary and discussion
TA2-2, Slide 2
The Basic Idea:

Reducing Costs Due to Maintenance
The overall cost per part goes way up due to the higher per unit cost of unscheduled Cost downtime
The benefit of lower scheduled downtime is Idealized Cap. Cap UnScheduled nearly wiped out by the UnScheduled UnSched led Downtime UnScheduled (nounscheduled downtime (due downtime) Downtime Scheduled Downtime to longer times for Downtime Scheduled Scheduled diagnosis), and diagnosis) and Downtimefault prediction in place D With f ti Downtime unscheduled Low due to conservative downs are Productivity approach Production reduced (turned intoto maintenance caused by high cost of y g Production Production scheduled) unscheduled downs More aggressive approach to maintenance improves Current Aggressive Optimized productivity slightly, Maintenance Strategy Practice Practice however
NSF Engineering Research Center for Reconfigurable Manufacturing Systems University of Michigan College of Engineering TA2-2, Slide 3
This is the opportunity for fault prediction
The Basic Idea:

Common Approaches To Scheduling PM
Manufacturers Estimates: Usually fixed times listed in the manual
Problems: Tend to be overly conservative, do not reflect factory conditions
E ti ti Estimating MTTF: Observing failures and adjusting time to failure MTTF

estimates; support part-count maintenance as necessary Problems: Does not account for variance in machine downtimes
Reliability Studies: Conducting independent statistical studies to model

reliability Problems: Expensive, requires factory downtime to conduct test
Failure Prediction through Data Corelation

Accounts for Variance Optimizes effectiveness O ti i ff ti Reduces Costs Can be implemented with event data
TA2-2, Slide 4
Project Objectives and Approach

Short-Term: Help Global Engine identify gaps in plant-floor systems
Better utilization of systems Improve data quality and data usability of these system
Mid-Term: Predict and reduce unscheduled downtime

Provide solutions for auto correlation of data sets
Long-Term: Schedule preventive maintenance through ECA rule based control Continuous: Provide and help implement best practices in data management for improvement in maintenance management and downtime prediction
Improving maintenance data quality Maintenance pooling Matching practice to specification Adjust project focus as necessary to adopt to the impact of changing economic conditions
TA2-2, Slide 5
Approach: Closing the Loop at Global Engine

ECA Control System Diagnostics Equipment/ Tool Control Maintenance Management Automatic Shutdown
Operations with Top Fault Count
60 120 100 80 60 40 20 0
Optimize Maintenance Scheduling
D e v ia tio n fro m A v e ra g e F a u lt C o u n t
50 40 30 20 10
Data Stores
Excel
C u m u la tiv e P e rc e n ta g e
OP300_1 OP110_2 OP60_3 OP160_3 OP280_3
Operations p
Engineers Machine Adjustment
BENEFITS Reduced Unscheduled Downtime Reduced Scrap
TODAY
OPC
PLCs CNCs Production Machines PLCs
Reduced MTTR
CNCs Production Machines
Improved Productivity
Production Machines
Production Machines
TA2-2, Slide 6
Outline
Introduction

Next Steps
TA2-2, Slide 7
Previous Deliverables
Look at main cost drivers of scrap, unscheduled tool changes and unscheduled maintenance Top Ten anomalies software installed at Global Engine and evaluated
Recognizing interesting events in process data Generate Excel report including Paretos
Lack of good data quality hurt effectiveness User interface design per Global Engine specifications somewhat
Matlab analysis module for maintenance event correlation installed at Global Engine and evaluated
Automated drill-down tool for maintenance investigation g C++ code auto-generated from MATLAB source no MATLAB license required
Best practices for improving maintenance management and data q quality y

Analysis of PM scheduling and reporting data quality Comparing maintenance practices to documented maintenance requirements
Student internships at Global Engine for Summers 06, 07 and 08

Top 10 Pareto Chart

Operations with Top Fault Durations O ti ith T F lt D ti
Devi iation from Average A To otal Dura ation (se ec)
25000 20000 15000 10000 5000 0 120 100 80 60 40 20 0
Screen shot of viewing the top ten anomalies

Faultiest Operations
Fault Count
P2 50 _1
Fault Total Duration
Operations

Deviation A from Average Fault Count t
Excel output
Fast learning rampup Ease of drill-down
60 50 40 30 20 10 0
P3 00 _1
P1 90 _2
P1 80 _4
P1 80 _8
Operation
120 100 80 60 40 20 0
OP300_1 OP110_2 OP60_3 OP160_3 OP280_3
Operations
Cumu ulative Perce entage
Cumu ulative Perce entage
Previous Result:
Strengthening Correlations with Normalized Overlay
A number of weaker Correlations can be grouped with overlays
After an overlay this operation demonstrates a gradual increase overlay, in faults over the week leading up to the unscheduled Maintenance. This can be used as a basis for predicting and rescheduling downtime downtime.
Outline
Introduction

Next Steps
TA2-2, Slide 11
Current Focus
Failure Prediction
Single fault type to maintenance g yp All fault types to maintenance
Leveraging improvements in data quality due to ongoing implementation of best practice improvements
Maintenance Under Low Production Conditions

Leveraging prediction information
Reducing Redundant Maintenance

Best practices for matching unscheduled and scheduled maintenance work Leveraging knowledge of unscheduled maintenance events in impacting scheduling maintenance
Failure Prediction Through Event Data:

Single Fault Type to Maintenance
Several Operations show strong correlations within the time scope examined, there are relationships that can be observed:
Continuous Reject Faults
Leveraging improved maintenance data quality through implementation of best practices Leveraging improved UMERC capabilities for analyzing per specific fault type, and for more automated analysis
NOSE PIECE WONT CLAMP TOGETHER @ 12:23
BOLTFEEDER NOT FEEDING BOLTS @ 18:21 BOLTFEEDER NOT FEEDING BOLTS @23:48 NOT FEEDING BOLTS @ 21:01 NOT FEEDING BOLTS ( y) @ 0:49 (Next Day)
BOLT FEEDER NOT WORKING @ 6:05 @ 15:39
@ 23:46 - Feed & Torque Station @23:46
It is not clear that these correlations will hold in the long term. However, if a relationship is confirmed, maintenance practices can be updated to reflect this knowledge The dataNSF Engineeringsufficient quality to where theseSystems is now of Research Center for Reconfigurable Manufacturing University of Michigan College of Engineering relationships can be confirmed our next step
TA2-2, Slide 13
Failure Prediction Through Event Data:

All Fault Types to Maintenance
Higher correlations (once verified) means that we can use the prediction to improve impact of scheduled maintenance
regression -7.545963e-001, mtype = AAA179608CPM type =grap all 200 180 160 140 120 100 80 60 40 20 0
BENEFITS Reduces the probability of unscheduled downtime occurrences Lower maintenance costs, higher MTBF, lower MTTR, more predictable machine operation
Faults on System Incur cost due to lost units, slowing production.
Alternative Schd. Monthly Downtime
Reduces the number of faults on the system, increasing system performance Allows maintenance scheduling to become adaptive
25 30 35 40
10
15
20
Schd. Monthly S Downtime NSF Engineering Research Center for Reconfigurable Manufacturing Systems University of Michigan College of Engineering
Unscheduled Downtime
TA2-2, Slide 14
Maintenance Under Lower Production Conditions:

Low Load Maintenance
In todays world factories are not operating at capacity

Focus is on reducing cost
However maintenance practices are often designed for atcapacity operation

Maintenance needs are often a function of production (e.g., part count) rather than time Maintenance practices are not adjusted when operating below capacity Money is being lost due to overly aggressive static maintenance practices
Solution: Implement part-count or event based maintenance part count event-based

Leverage part tracking and maintenance prediction Leverage fact that maintenance system supports part-count based maintenance triggers
Issues
Part tracking insufficient to support reliable count-based triggering Maintenance prediction needs to be verified with new data
Maintenance Under Lower Production Conditions: Low Load Maintenance

Example: Since the fault count on several operations track upward ahead of a maintenance operation as a result of part count, the prediction of the event can be used to trigger maintenance, resulting in longer times between p maintenance in low load production conditions
High Fault Rate tracks System Load
BENEFITS This alternative leverages existing information on the system It doesnt require an update to the part tracking system d t i d t t th t t ki t The maintenance system can support event-based triggering
ISSUES
Need to verify causal relationships with newer higher quality data y p g q y
Reducing Redundant Maintenance (1)

In a typical time-based maintenance system, a sudden failure can trigger an unscheduled maintenance event Schd. PM Schd. PM
Schd. PM
Failure!
time
TA2-2, Slide 17

If the unscheduled PM work is matched to the Scheduled PM, the PM schedule, if left unchanged is redundant and non-optimal Unscheduled PM Schd. PM Schd. PM
Schd. PM
Redundant; mis-timed
time
TA2-2, Slide 18

The maintenance scheduled should be reset, resulting in fewer maintenance events, lower downtime, and lower maintenance costs Unscheduled PM Schd. PM Schd. PM
Schd. PM
KEYS TO MAKING THIS WORK Linkage between unscheduled maintenance events and triggers in DataStream Matching Unscheduled to Scheduled Maintenance work orders p best practices
time
Datastream Counter Reset
TA2-2, Slide 19
Other Ongoing Efforts of Interest:

Unified Data Layer
Effort to integrate various data systems to increase the accuracy of regression analysis. An audit of all databases was performed to evaluate data quality and possibility of creating a unified data layer.
Global Engin Databases
BENEFITS M E T A D A T A L A Y E R Allows for a more comprehensive visualisation and understanding of d t di f factory data.
ActivPlant
DataStream ataSt ea
x x xx x x x x xxxxxxx xxx xx x x x xx xx xx xxx xx

Correlation across multiple variables
Facilitates correlation analysis across many variables i bl

Engineer
MPTS
Allows islands of automation to be focused on the same factory objectives ISSUES * Poor data quality prevents unification of data.
TA2-2, Slide 20
Scrap Data
Other Ongoing Efforts of Interest:

Improving Best Practices on Test Stands Pass Engine Tear Down
Update the Book of Knowledge g
Test Stand
1.) No Problems in the Engine (Engine Passed) (90%) of time 2.) Minor Problems that can be fixed at cold stand (very rare) 3.) Major Problems, Engine needs to be Torn Down a.) O ti ) Option 1: problem exists i th B k of k 1 bl i t in the Book f knowledge, and l d d root cause can be identified. Result: Fix Operation Identified. b.) Option 2: problem's root cause cannot be identified. Engine is not immediately torn down due to lack of time and resources. resources c.) Option 3: Engine is immediately torn down and operation may halt in order to determine root cause.
Operation Operation Operation Operation Two One Four Three KEYS TO IMPROVING BEST PRACTICES The main decision that needs to be made is whether a reject that has not been observed is worth examining the root Reconfigurablethe problem causes of Manufacturing Systems NSF Engineering Research Center for TA2-2, Slide 21 University of Michigan College of Engineering The Largest Area of improvement is to determine a way for modeling the Likelihood of a reject seen occurring again
Outline
Introduction

Next Steps
TA2-2, Slide 22
Next Steps:
Failure Prediction
Project has shifted focus somewhat in light of the current economic conditions
Focus on reducing cost NOW Ideas for reducing cost in light of operation under capacity
Data Quality best practices improvements are starting to pay off

We have evidence of correlations between faults and maintenance events in the latest data We need to verify these correlations in new data coming in Unfortunately older data (pre-best practices improvements) is not that useful for verification Unfortunately some of the newest data reflects operation much under capacity We have to decouple the capacity issue to verify correlation
TA2-2, Slide 23
Next Steps:
Low Production Load Maintenance
Correlation, if verified will allow for implementation of event-based maintenance
Newer data (reflecting newer best practices) should be of sufficient quality to (1) verify correlations, and (2) determine if maintenance is a function of part count as opposed to (or in addition to) time
Implementation of improved part tracking and linking to p p p g g maintenance for part-count based maintenance scheduling is a longer term issue Longer term, all three types of maintenance scheduling term should be in place, based on prediction information (or lack thereof)
Time-based maintenance Part-count based maintenance Event-based maintenance Combinations
Next Steps:
Reducing Redundant Maintenance
Global Engine is aware of what needs to be done Fix is largely a combination of engineering and best practices improvement at this point
Linking of unscheduled corrective maintenance events into the preventative maintenance system Aligning the work order descriptions of unscheduled maintenance operations with scheduled maintenance operations
TA2-2, Slide 25
Next Steps:
ROI Calculations
Impact of current efforts should be easy to convert into ROI numbers
E E.g., reduced number of maintenances per unit time X cost of d d b f i t it ti t f maintenance event in terms of man hours and consumables We hope to use ROI data to drive additional improvements in integration and best practices i t ti db t ti Improved part tracking / counting Aligning of unscheduled and scheduled downtime work order descriptions Resetting capability for maintenance schedules
TA2-2, Slide 26
Next Steps:
Other Efforts
Data consolidation / meta layer Improve best practices on test stands; decision process with rejects; analysis of continuous data E l Explore i incorporating MPTS d t i t model f ti data into d l for Unscheduled Downtime Prediction C ti Continue t impact implementation and refinement of to i ti l t ti d fi t f best practice improvements
Data quality improvements to support prediction, consolidation and ROI goals
TA2-2, Slide 27
Summary
Implementation of best practices for improved data quality are starting to pay off
We are seeing potential correlations between faults and maintenance events Signal could be strong enough for prediction
Current focus areas reflect economic environment

Fault p ed ct o au t prediction maintenance sc edu g a te a ce scheduling Low production load maintenance scheduling improvements Reducing redundant maintenance
Next steps focus on verifying analysis with new data and providing justification for best practice improvement investment through ROI analysis
Discussion
Need to determine the type and level of longer term support for the project
No Summer internship in 2009 Will likely have close interaction with Global Engine via regular visits
Need to determine if there is reusability

Elsewhere in Chrysler Other ERC members
If there is a desire to close the project, 4 6 project months additional to tidy up deliverables would be ideal
There is still research to be done Prioritization with other projects
Questions?
Backup Slides
TA2-2, Slide 30
Examples of Data Quality Issues Encountered

System collects only a subset of the data generated from the PLCs. There is often not enough information to identify strong correlations to support control. Historical Data is limited, making larger trends and behavior difficult to identify. Maintenance records have missing data. Insufficient standardisation of data, manually entered data , y unsuitable for computerised analysis. Scrap code definition process results in unbounded growth of codes nearly useless for correlation analysis
Ad-hoc scrap code creation
Number of Scrap S Codes s
Time
Samples of Recommendations for Improving Data Quality

Modular and Hierarchical Reason Code Scheme
Incorporate an ID system for maintenance to help record keeping and reduce redundancy E.g., Maintenance and scrap databases
System Wide Data Unification

Create a factory meta-data layer for access analysis and drill-down meta data access, drill down Create a unified labeling system so that a part can be tracked through the system.
Extended Access to Historical Data

Build in a mechanism for access to large vectors of archived data.
Deep Analysis Of Key Operations

Select a single operation within the system and analyse its behaviour over a long historical period to understand fault alarms and downtime behavior Use this process to better identify underlying data quality issues
ROI of Scaling Maintenance to Factory Load

While there may be additional costs to implement a scaling system; the reduction of costs due to unnecessarily performed downtime indicate a high cost benefit ratio.
BENEFITS INVESTMENTS A system by which to track parts. --OR OR A mechanism for scheduling downtimes w.r.t fault data. R d Reduced costs due to d t d t un-necessary maintenances on the system. Increased Production time. The ability to cope p variable production situations.
TA2-2, Slide 33
Dataflow to Assist Analysis

Historical Database MS Access (Manual) ActivPlant Scrap GEMA Int. Maint. PPS/PDCA
D e v ia tio n fr o m A v e r a g e F a u lt Count
60 50 40 30 20 10 120 100 80 60 40 20 0
Predict Unscheduled Downtime
ActivPlant Maint. M i t CM, PM DataStream
Excel

60 120 100 80 60 40 20 0
D e v ia tio n fro m A v e ra g e F a u lt C o u n t
40 30 20 10
Maint, JavaScript faults based on duration d ti

Unsch. Tool Change.
MATLAB (C++)
Graphs, Correlations
0
C u m u la tiv e P e rc e n ta g e
50
30 O P 0_1 11 0 OP _2 6 O P 0 _3 16 O P 0_3 28 0_ 3
Machine Faults
Number, Number JavaScript duration of faults
Graphs, Graphs Correlations

0
OP
Operations
C u m u la tiv e P e r c e n ta g e
OP300_1 OP110_2 OP60_3 OP160_3 OP280_3
Operations
Analysis
Machine Reject MPTS
Plan: Put machine faults together with machine

rejects; incorporate information from Unschd. Tool Change and Scrap data if possible. (see dotted line)
TA2-2, Slide 34
Excel (Manual)
Excel (Manual)
UML DATA LAYER (CURRENT)
TA2-2, Slide 35
NSF Engineering Research Center (ERC) for
Reconfigurable Manufacturing Systems (RMS)
Cyclic W C li Waveform Signal Analysis for Monitoring f Si l A l i f M it i & Diagnosis of Powertrain Manufacturing Systems
Research Team: R hT
ERC/UM: Judy Jin, Kamran Paynabar, Yong Lei, Qiang Li, GM: John Agapiou (R&D); Ed S Sponseller & St ll Steven N Norman (GMPT Li (GMPT- Livonia); i ) Thomas Gustafson & Phillip Steinacker (GMPT-Pontiac) Chrysler: James Wang, Eugene Kuo, Mark Skelly, John Gartner
Presenter: Judy Jin
March 16, 2009
#1
Outline
ProjectOverview&ProcessBackground Accomplishments & Tech Transfer Plan at GM Accomplishments&TechTransferPlanatGM NewinvestigationsatChrysler
Comparison of data collection between Chrysler and GM ComparisonofdatacollectionbetweenChryslerandGM AnalysisofavailableoriginalwaveformsignalsatChrysler
New project proposed by GM NewprojectproposedbyGM Milestone&FutureWorks
#2
Project Overview
Problem
Cyclic waveform signals are widely used for online monitoring of powertrain manufacturing process. The trial and errors approach for features extraction & monitoring limits may not always be effective.
Goal
To develop systematic waveform signal analysis methods to improve online monitoring systems for powertrain manufacturing processes.
Deliverables and benefits

Develop a generic signal analyzer for a broad class of cyclic waveform sensing signals to D l i i l l f b d l f li f i i l t automatically set up monitoring limits to improve first time quality and reduce ramp up time for new production lines; effectively extract monitoring features from online sensing signals to reduce both false rejects and miss detection for reducing mfg cost; continuously learn and enhance diagnostic capability to quickly identify the root causes for reducing defects & downtime.
Main tasks
W k with engineers at GM (&Ch l i this quarter) t collect production/DOE data from the Work ith i t (&Chrysler in thi t ) to ll t d ti /DOE d t f th selected process; Develop data analysis algorithms for data preprocessing/signal alignment/signal segmentation /online monitoring charts to characterize/monitor process operation states; Characterize process fault patterns to enhance diagnostic capability; p p g p y; Software development, plant testing, and validation.
NSF Engineering Research Center for Reconfigurable Manufacturing Systems University of Michigan College of Engineering #3
ProcessBackground
Valve Seat Pressing Machine
Sensor data Sciemetric System Problem: A high false reject rate is a top concern. Goal: Improve production throughput by effectively using online sensor monitoring systems. i i Recorded manual inspection 28%
4000 Peak Force
3000 Loa ad
True Detection65%
2000
Gap
1000 Force Work 0 -0.25 -0.2 -0.15 -0.1 Distance Depth p -0.05 0
Unrecorded inspection72%
Reduced to 5% by ERC
#4
Overview of Proposed New Monitoring Methods

High dimensional non-stationary sensing signals Cyclic waveform signal analyzer
6000
Low dimensional physical interpreted features SL
Monitoring control limits
Force Root cause: misaligned signals.
Alignment algorithms Segmentation algorithms Feature extraction algorithms Online monitoring limits Fault classification
5000
false reject
4000
3000
2000
1000
LVDT
-1000 -0.3 -0.25 -0.35 -0.2 -0.15 -0.1 -0.05 0 0.05
Multiscale wavelet analysis M lti l l t l i

yk h 1 d1,k
di,k
y y
New aligned depth by UM
Sciemetric depth
l1
c1y,k
...
ci1,k
h i ciy,k
djy0,k
li
...
cy01,k j
hj0 cy0,k j
lj0
#5
Accomplishments & Tech Transfer Plan at GM

Major Accomplishments:
ERC-team has proposed new algorithms for false reject reduction, the validations show: UM algorithms can help reduce false rejects from the current 35% to 5%; UM algorithms have the potential benefit to reduce miss detection of bad products by setting more effective specification limits.
Current Tech Transfer Status:

GM is interested in implementing new algorithms and initiated efforts for the ERC/UM team to work with the Sciemetric Company. Technology transfer agreement was made between ERC, GM and Sciemetric Company. p y In this quarter, UM has provided the program codes to Sciemetric Company, which will be implemented into GM monitoring system for further validation.
#6
Newinvestigationsat g Chryslerinthisquarter
#7
Comparison of Data Collection Schemes Between Chrysler and GM

GMPTatLivonia Accepted parts
Originalsensingsignals Scimetricfeaturesdata Scimetric features data
Every100 parts Notsaved Not saved
ChrysleratMac1 Accepted Rejected parts parts

Last25parts Last25parts Allparts All parts Allparts All parts
Rejected parts
Allparts Allparts All parts
Two analyses have been done based on available original waveform signals: y g g For the rejected samples having regular signal profiles (8 samples), are they false rejects? For the rejected samples having irregular signal profiles (17samples), what are the root causes?
#8
Analysis1:
InvestigatewhetherSciemetricrejectedsampleswith regularsignalprofilesarefalserejectsornot(8samples).
#9
RejectionsBasedonOldSciemetricDepth
UL=0.5
(in)
Parts are rejected by the Sciemetric depth feature. (suspect of false reject)
LL=-0.5
#10 10
SignalComparisonbetweenRejectedandNonrejectedParts
10000
Specification Limits Blue: 25 samples of accepted parts by Sciemetric
8000
Red: 8 samples rejected by Sciemetric (suspected false rejects)
Force (lb bs)
6000
4000
Comment: Misaligned signals may cause false rejects g g y j based on Sciemetric depth feature.
2000
-2000 -35
-30
-25
-20
-15
-10
-5
LVDT (in)
NSF Engineering Research Center for Reconfigurable Manufacturing Systems University of Michigan College of Engineering # 11
NewFeatureAlignedDepthafterSignalAlignment
Algorithm of calculating aligned depth: 1. Use wavelet analysis (Harr transformation) to find contact point. Contact point: The point where the press tool actually contacts the part leading to the change point of force signals. g g p g 2. Calculate new feature aligned depth. The difference between the contact point and the maximum LVDT value, which reflects the actual moving range.
4500 4000
-------
Out-of-control In-control
4500
4000
3500
3500
3000
3000
2500
2500
Aligned by contact point
2000
2000
1500
1500
1000
1000
Contact point
500
500
-500 -0.25
-0.2
-0.15
-0.1
-0.05
0.05
0.1
-500 -0.1 01
-0.05 0 05
0.05 0 05
0.1 01
0.15 0 15
0.2 02
Misaligned Signals
Aligned Depth
# 12
SignalsAfterAlignment
10000
8000
Blue: 25 samples of accepted parts by Sciemetric
Force (lbs)
6000
Red: 8 samples rejected by Sciemetric (suspected false reject rate based on 25
4000
samples=8/25=32%) Comment: Aligned depths are
2000
within the limits.
-2000 -25
-20
-15
-10
-5
10
LVDT (in)
MonitoringBasedonNewAlignedDepth
UL=6.93
(in)
LL=5.93
#14 14
RequestofNewDataCollectionfor FurtherValidationatChrysler Further Validation at Chrysler

1. Record the original sensing signals for every 100 parts to verify the th new depth f t d th feature; 2. Record quality inspection data to verify rejections.
#15 15
Analysis2:
ClassifyIrregularSignalProfilesofRejectedParts Cl if I l Si l P fil fR j dP (17samples)
# 16
Irregular Shape Signals at Chrysler: Missing seats

-5 -10 10 -15 Force e LVDT( (lbs) (in) -20 -25 -30 -35 0
Force (lbs) e 10000 8000 6000 4000 2000 0
500 Time (ms)
1000
1500
-2000 0
200
400
600 800 Time (ms)
1000
1200
1400
10000 8000 6000 Force (lbs) 4000 2000 0

NSF Engineering Research Center for Reconfigurable Manufacturing Systems University of Michigan -2000 of Engineering-25 College -35 -30 -20 -15 -10
-5
# 17
LVDT (in)
Irregular Shape Signals at Chrysler: Interrupted operations

-27.5
120 100
-28 28
80
LVD (in) DT
Force (lbs) e
200 400 600 800 Time (ms) 120 1000 1200
-28.5
60 40 20 0 -20
-29
-29.5
-30 0
-40 0
200
400
600 800 Time (ms)
1000
1200
100 80 Force (lb bs) 60 40 20 0 -20

NSF Engineering Research Center for Reconfigurable Manufacturing Systems -40 University of Michigan College of Engineering -30 -29.5 -29 -28.5 -28
-27.5
# 18
LVDT (in)
Mapping Table Developed based on GM Data ___ Example of classified irregular profiles at GM
Faulttype LVDT (springproblem) LVDT
Highly oscillated LVDT HighLVDT Normal N l
Force
Force Highly oscillated (cableproblem) OscillatedLVDT LVDT &LVDT & LVDT (highvariance) (hi h i ) (springproblem)
Normalshape p
No correct reading
MissingPart
Normal Normal
No local force High value of peak
# 19
Improved Diagnostic Decision for Monitoring System

Incoming SensorSignal
Detect & classify Irregular signal profiles?
Improvement by UM To Add a detection& classification module. Developed better monitoring features (e.g. aligned depth)
No
Providefaultreport& correctionaction
Yes No
Calculatemonitoring features
Outof specification limits(SL)?
Yes
Confirmfaultdetection
# 20
NewProjectProposedbyGM __ImproveLeakingTest
# 21
NewprojectatGM:ImproveATCLeakingTest
The Whole Leak Testing System
# 22
New Project Activities with GM

Site visit: UM and GM teams visited Advanced Test Concepts (ATC) Inc. in Indianapolis on Feb 9&10; Identify new projects: Project 1: Signature analysis and self-learning of fault patterns for abnormal detection and root cause identification . Project 2: Verification of the adaptive test strategy to reduce the leak test cycle time and improve the production throughput. Project 3: Development of the standardized procedures for self-calibrating test tools at the tool setup stage for reducing the setting up time and engineers trial and error calibration efforts. Next step Work with the GM team to identify a plant, which can be used for the data collection and method/tool development in the next step.
# 23
Milestones and Future Plans

Last Year
Defined project scope and candidate manufacturing processes. p j p gp Collected production data at GMPT plants at Pontiac and Livonia. Understand current practice and the GM need. Conducted experimental tests and analyzed DOE data. Developed signal alignment algorithms using Wavelets. p g g g g Found the root cause of high false rejects of Sciemtetric system. Developed new algorithms for reducing false rejects. Prove the concept of profile feature extraction for gap detection. Collect more data for further validation of new features and signal alignment algorithms. Integrating our proposed alignment algorithm with Scimetric software. Propose an algorithm for determining specification limits. Analyze irregular fault patterns to enhance diagnostic capability for detecting and classifying sensor problems and missing part. W k with Sciemetric company for software implementation. Work ith S i ti f ft i l t ti Work with Chrysler for algorithms validation and implementation Identify the new projects for improving leaking test at GM Continue validation analysis at Chrysler. Work with GM for the new projects of improving leaking tests.
June 2008
Dec 2008
Next Quarter
Next year
Thankyou! Thank you! Q&A Q&A
# 25
3.AnalyzingSciemetricfeaturedata 3. Analyzing Sciemetric feature data correspondingtoacceptedparts
# 26
Depth
0.4 0.2 Dept th 0 -0.2 -0.4
Featuresboxplotovertime
Work
12000 10000 8000 6000 4000
1 2 3 4 5 6 8 9 10 11 12 15 16 17 18 19 29 30 36 37 38 39 40 43 44 45 46 47 49 51 52 53 56 57 58 59 63 64 65 66 67 70 71 72 77 78 79 80 81 84 85 86 87 88 Date index
Peak
10000 9000 8000 7000
1 2 3 4 5 6 8 9 10 11 12 1 16 1 18 19 29 30 36 3 38 39 40 43 44 4 46 4 49 51 52 53 56 57 58 59 63 64 6 66 6 70 71 72 77 78 79 80 81 84 8 86 8 88 15 17 37 45 47 1 2 3 6 8 9 65 67 0 1 2 8 9 85 87 Date index
Force
3000 2000 1000
1 2 3 4 5 6 8 9 10 11 12 15 16 17 18 19 29 30 36 37 38 39 40 43 44 45 46 47 49 51 52 53 56 57 58 59 63 64 65 66 67 70 71 72 77 78 79 80 81 84 85 86 87 88 Date index
1 2 3 4 5 6 8 9 10 11 12 15 16 17 18 19 29 30 36 37 38 39 40 43 44 45 46 47 49 51 52 53 56 57 58 59 63 64 65 66 67 70 71 72 77 78 79 80 81 84 85 86 87 88 Date i d D t index
Sep 08
Oct 08
Date index
Nov 08
Dec 08
# 27
Depthboxplotovertime(goodparts)
0.4 0.3 0.2 0.1
Sep 08
Depth h Depth
0 -0.1 -0.2 -0.3 -0.4 -0.5 1 2 3 4 5 6 8 9 10 11 12 15 16 17 18 19 29 30 36 37 38 39 40 43 44 45 46 47 49 51 52 53 56 57 58 59 63 64 65 66 67 70 71 72 77 78 79 80 81 84 85 86 87 88
Oct 08
Nov 08
Dec 08
An obvious shift in depth from Oct 08 Possible reason:

Change in cylinder head raw part Resetting the production process
Date index
Date index
FeaturesScatterMatrix(goodparts)
# 29
FeaturesScatterMatrix(OcttoDec)
Low peak and low depth
Low Peak and Center for Reconfigurable Manufacturing Systems NSF Engineering Research University of Michigan high force College of Engineering
# 30
Projectedfeatures (OcttoDec08)
All data correspond to good parts (based on Sciemetric report)
Although all parts are good, they can be g p g y clustered in two classes by using features. To understand the reasons further study is needed.
10000
8000
Pink: part # 2 which is close to upper limits

6000
Force(lb bs)
Blue: other accepted parts
4000
2000
1800
1600 1400
-2000 -7
1200
-6
-5
-4
-3
-2
1000 800 600 400 200 0
-1
LVDT(in)
-200 -7 -6.8 -6.6 -6.4 -6.2 -6 -5.8 -5.6 -5.4 -5.2
# 32
-5
8000 7000 6000 5000

Force(lbs s)
Black: rejected parts close to the lower limit Red: other rejected parts
4000 3000 2000 1000 0 -1000 -7 7
-6 6
-5 5
-4 4 -3 3 LVDT(in)
-2 2
-1 1
# 33
AlignedSignals
10000 8000
Blue: accepted parts

6000 Force (lbs )
Black: rejected parts close to the lower limit Red: other rejected parts
4000
2000
-2000 -1 1
3 4 LVDT (in)
# 34
Detecting and Classifying Sensor Failures ---- Proposed Mapping Algorithm by ERC/UM
Faulttype
TypeILVDT (springproblem) (spring problem) TypeIILVDT
LVDT
OscillatedLVDT (highvariance) (high variance) Stepshape (highlypositiveslope) Spikeshape (highlypositive& negativeslope) Normal Nocorrectreading/No slope HighLVDTreading
Force
Normal Normal
TypeIIILVDT TypeIVforce (cableproblem) TypeVforce&LVDT
Normal eitherverylowforceor ith l f overpeakforce Nocorrect No correct reading/Oscillatedforce (highvariance) Nocorrect reading/Oscillatedforce (highvariance)
# 35
TypeVImissingSeat
SignatureOutputs
Quick Fill+Fill
Stability St bilit
Test T t
Master Part + EC
Master Part Select limit to be used for leaking engines.
# 36
IdentifyRootCauseofHighFalseRejectRate
6000
Force
5000
Blue: classified as a good part by Sciemetric Red: false rejected as a bad part by Sciemetric
SL
4000
Both signals seem to be good parts.
3000
2000
Rootcause:misalignmentonsignals.
1000
-1000 -0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0.05
Sciemetric depth is based on LVDT absolute value, which is not a true tr e reflection of the act al mo ing range actual moving
Depth p
Depth p
# 37
NSF Engineering Research Center for
Reconfigurable Manufacturing Systems
Statistical Output Analysis of Steady-State p y y Simulation Models In SimuVeri Software

Sam Yang, Wencai Wang and Jack Hu (UM) Susan Ostrowski and Annette Januszczak (Ford) March 16, 2009
The University of Michigan, Ann Arbor
Outline Project Overview j SimuVeri System Architecture Introduction to Statistical Output Analysis Application Case Whats Next
#2
Project Overview
Background: The project arises from an earlier Ford-ERC project at CEP
The manufacturing processes and models are very complex; Model verification and experimenting are time consuming; Models are not completely tested on system level before experimenting; Resulting in exaggerated throughput and poor model applicability.
Goals:
To develop Computer Aided Testing (CAT) tools for automatic error checks; To provide user friendly software platform supporting various validation, experimentation and optimization tec ques techniques.
Architecture of Application Software

Users select testing strategies Software executes tasks & outputs Results
Engineers develop model d l d l
WITNESS runs each scenario

Introduction of Statistical Output Analysis

Random inputs/parameters in the simulation models result in random observation outputs (performance measure estimates) with some distribution. Questions about experimenting simulation models.
How do we start the simulation (warm-up)? How long should the model be run (how much simulated time before stopping the run)? How many samples of the performance measures should be collected (how many replications)? H How should th output b analyzed? h ld the t t be l d?
#5
Steady State Simulations
#6
Steady State Simulations

The system has reached steady state where the performance is independent of the initial conditions conditions. Examples
Production line simulations the line starts where it left off at the end of prior shifts. Emergency rooms. Airplane scheduling at airport.
#7
Steady State and Running Time

How to start the simulation?
Typically a warm-up period is used to minimize any impact of initial conditions. How long should the warm-up period be? warm up How long to run the simulation?
Correlation Analysis
#8
The Number of Replications

Simulation models are used for experimentation. p
One simulation replication a single sample (realization) of each system performance measure.
n independent replications n independent samples from the

same distribution.
Confidence Interval Analysis
#9

Consider a single performance measure Let Xi be the measure. random variable that represents the value of the performance measure for the ith simulation replication. xi = outcome/realization of Xi from the ith simulation replication. Since the Xi are independent and identically distributed random variables th performance can b characterized d i bl the f be h t i d using the typical confidence interval.
# 10
Analysis of Output (1)

The approximate (1 ) *100% confidence interval
s x z1 / 2 * n
Central limit Theorem: regardless of the distribution of (n each Xi ,X ( ) will b approximately di t ib t d as a h ill be i t l distributed normal random variable when n
# 11
Analysis of Output (2)

The approximate (1 ) *100% confidence interval
s x t n 1,1 / 2 * n
Assumes the sample average is from a normal distribution. di t ib ti
# 12

How to estimate number of independent replications required for a desired precision. The half-width h of this confidence interval is
h = t n 1,1 / 2 * n=t
2 n 1,1 / 2
s n s2 for a desired h (precision level). 2 h
# 13

Substitute
z1 / 2 for t n 1,1 / 2 s2 n z12 / 2 * 2 h

Use this formula to approximate the number of replications needed to get a desired half-width (precision) for some half width performance measure.
# 14
An Application Case
# 15
An Application Case
Time 1.484055403 1.931330511 3.498247812 6.424550926 24.20430586 33.38210516 33.78572903 46.44300943 47.72927941 47 72927941 60.97070454 80.76142078 Observation 1.484055403 0.200569173 0.416304048 2.639221143 6.142266534 5.722407758 5.420538552 2.564226808 0.695810334 0 695810334 12.17619063 30.96875537
Avg Stddev
8.187022 5.428285
# 16
An Application Case
z1 / 2 for t n 1,1 / 2 nz
2 1 / 2
s 2 (AvgTIS ) * h2
8.187022 5.428285
z10.025 = 1.96 5.432 n 1.96 = 453 2 .5

2
Avg. Stdev
# 17
Whats Next
LITERATURE REVIEW AND PROJECT DEFINITION
Project needs and direction Verification and validation approaches
WITNESS COMMUNICATION PROTOCOL

Witness file parser Witness file generator Interfaces
VERIFICATIONMODULE
Code the module
Specific errors to be detected
Verificationstrategies Codethetestingstrategies
VALIDATIONMODULE
Validation needs Validation strategies Code strategies
OPTIMIZATIONMODULE
Code and test optimization Optimizationneeds methods Exploreoptimizationmethods Completed InProgress FutureWork
# 18
Engineering Research Center for
In-line Inspection of Engine Valve Seats

Dr. Dr Reuven Katz and Sankalp Arrabolu
March,13th, 2009
In-line inspection of engine valve seats

Seat angle Gage
Seat length
Valve guide
Deck seat (0 1o) throat angles Deck, (0.1 ), Seat length Seat roundness at gage Seat runout at gage wrt valve guide
Project Goal
In-line measurement of valve seat geometry (cycle time ~ 45 seconds) Rapid and accurate non-contact measurement Measurement of seat angles and seat length Preliminary repeatability test Evaluate in-line application feasibility Comment: All the study was done without having a defined specification of the problem
Approach 1: Single Cross Section
Approach 2: Least Squares Cone Fit w/ Five Cross Sections
Seat angle
ERCResults CMMResults Run1 Run2 Run3 Run1 Run2 Run3 60segment angle SeatAngle Seat Angle 30segment angle SeatLength 60.53 59.39 59.99 45.18 45.18 44.98 45 18 45 18 44 98 30.28 30.33 30.27
1.7137 1.6979 1.6991 1.729 1.772 1.738
45.15 45.34 45.18 45 15 45 34 45 18
Valve seats R&R test results

Resultsoftherepeatabilitytest
performedfor50measurementsacross5crosssections
SeatAngle (degree)
DeckAngle (degree)
ThroatAngle (degree)
Roundness (mm)
SeatLength (mm)
GageDepth (mm)
Average
45.07468
60.40652
30.34172
0.02988506
1.747624
12.62412
Standard Deviation
0.017752769
0.043423985
0.017123847
0.002473562
0.012789657
0.002429748
Accomplishments and Next Steps

Accomplishments:
Angle measurement to within 1 degrees achieved using both approaches h Seat length measurement to within 0.2 mm students. Two axis demonstrator designed and built by ME450 students Complete statistical analysis of cone-fits for improving accuracy. Design repeatability set-up and testing
Next Steps:
Evaluate the implementation feasibility i e increase measurement i.e. speed, optimize data collecting path Test serial robot and a evaluate the use of a PKM
Operational Data of the Current Measuring System
Optimet Conoprobe Laser Scan Frequency : The Motion Stage Forward Motion Maximum Speed Possible : y (p y p ) Current System (presently not optimized for time) Forward Speed used : Backward Speed used : Stop time between scans : p
3000 KHz
5000 mm/min 100mm/min 1000 mm/min 1 sec
Methods for Time Optimization with above Specifications
Current Method Time Taken = 193 sec
Time Taken= 176 sec
Time Taken = 88sec
Challenges
To find the minimum time : we need to increase the speed of th scan. d f the Increasing speed reduces the number of data points captured. captured The Speed Vs Data Capture is an OPEN ISSUE. Need to decide which parameter to compromise on for on, the best time optimized performance.
11
Parallel Kinematic Mechanism (PKM) for Precise Location of Optical Sensors P i L ti f O ti l S

Dr. Hagay Bamberger Dr. Reuven Katz Date: 3/16/2009
The Goal & The Method

Develop & build a PKM demonstrator, which will precisely locate optical sensors like camera or laser sensors, Use PKM for Valve Seat measurements as well as for Small Bores inspection projects p p j The suggested PKM possesses 4 degrees of freedom that are required for locating precisely optical sensors, within a desired workspace The advantages of a PKM: A Accuracy High rigidity Large payload capacity Fast dynamic response
Description of the Suggested PKM

The mechanism consists of: M i Moving platform capable of 4 d l tf bl f degrees of f d f freedom: two translations and two rotations 4 linear motors on the base
Laser Motionstage
SightPipe
Project Layout
Analysis & Synthesis: Ki Kinematics, D ti Dynamics, W k i Workspace, Si Singularities, l iti Structure, Joints Detailed mechanical design Detailed control design Building and calibrating Tests: Accuracy, Repeatability, Required resources: CAD designer Control expert Budget of ~$30K for hardware
Estimated Timetable for PKM project

Starting: 4/1/2009 Mechanical design review: 5/2009 Control design review: 7/2009 Component purchasing & manufacturing: 9/2009 Working prototype: 11/2009 Tests: 12/2009
Internal Thread Measurement

Dr. Reuven Katz, Dr. Hongwei Zhang and Dr. En Hong g g g
Mar. 16, 2009
Project Overview
Goals:
Develop methodologies for the inspection of geometrical features of internal threads in machined automotive parts.
Deliverables and benefits

The two methods to be presented enable in-process internal thread quality erification sing q alit verification using optical sensors. sensors The approaches allow to extract thread pitch, major and minor diameter, flank angle and even the starting point of the thread with respect to a reference location on the perimeter. perimeter
Main tasks
Laser scan measurement using Optimet sensor Optical inspection using a CCD camera with sightpipe R&R test to be done partially
Methodology using a laser sensor and set-up

Principle
InternalThread
Rotary Rotary Motion OpticalS ensor Z Y X MotionStages Motorized periscope
Probe
Mirror
Zaxis
Xaxis RotaryStage Y axis

MotionStages
Rotary Motion
Method:
Periscope
Measuring internal threads using a Laser Range Finder (Optimet Sensor) integrated with a motorized periscope designed at ERC.
45Mirrorinside StepperMotor Sensor
Measurements results of M12X1.25 internal thread

STANDARD: ANSI/ASME B1.13M-1983 (R1995) M12 X 1.25 Unit: mm Minor Diameter --D1 Min. Min Max. Max 10.647 10.912 Pitch Diameter --D2 Min. Min Max. Max Tol. Tol 11.188 11.368 0.18 Major Diameter --D Min. Min Max. Max 12 12.360
Measured Parameters 0 1.248 0.639 180 1.251 0.696 12.014 10.667 10 667 90 1.252 0.539 12.127 10.871 10 871 270 1.246 0.615 Average 1.249 0.622 12.071 10.769 10 769 Minor Dia. Major Dia.
Pitch (mm) Height (mm) Major Dia. (mm) Minor Dia (mm) Dia.
Pitch
Conclusion 1:
Compared to the standard data, all the parameters we get are within acceptable limits for the designated thread type.
axial cross section
Measurements results of starting location of a thread

Method:
Measuring the axial distance between the edge of the threaded bore and the center point of the bottom of the first thread tooth. Four measurements were taken at four different angles. The area of thread started is determined by comparing these four values. Theoretically, the largest and smallest must be neighbors and the starting point of the thread i l t ti i t f th th d is located b t t d between th them.
90 =2.288 Area of Thread Started
II I
Conclusion 2:
* The starting point of the thread is located in quadrant IV. * The helix is clockwise.
180 =1.942
III IV
0 =2.531
270 =1.384 radial cross section
Methodology and setup based on a CCD and a Sight-pipe

360 degree view from the sight pipe Selected annular zone Reconstruct final image
Lens& Illumination
Define the annular Zone of each Frame
Extract and unwrap the annular Zone of each Frame
Stitching Line strip
CCD
Sight Pipe
360-degree-view Line Scan Flow

Tilt Stages Motionstages LED Illumination
Internal Thread Measurement System
Conical Lens
Optical principle of the sight-Pipe The integrated sensor
Measurements results
1. A smoothing filter followed by Prewitt filter is used to bring out edges.
2. LabVIEW Shape detection is used to extract average angle of threads. 3. A series of line profiles are generated perpendicular to the thread lines and peaks are detected.
Discussion on Method 2 (3D Digitalization)
Challenges on 3D Digitalization
Phase shifting could be achieved by using Light Modulation Technique. With digital light source we can fulfill the phase shift. How to find the Sensitivity factor by changing the cylinder diameter as the reference plane moves. The angle of the illumination and optical axis is either very small or immeasurable which could cause K unsolvable ( we may start with flat surface calibration. ) The misalignment of the sight pipe system and surface quality variation may lead to errors in phase shifting measurement.
Next step: Prove if it is doable or not.
The End
10
Discussion on Method 2 (3D Digitalization)
11
Challenges on 3D Digitalization
Phase shifting could be achieved by using Light Modulation Technique. With digital light source we can fulfill the phase shift. How to find the Sensitivity factor by changing the cylinder diameter as the reference plane moves. The angle of the illumination and optical axis is either very small or immeasurable which could cause K unsolvable ( we may start with flat surface calibration. ) The misalignment of the sight pipe system and surface quality variation may lead to errors in phase shifting measurement.
Next step: Prove if it is doable or not.
12
Outlier Teeth Detection in Sprockets p

Dr. Dr Reuven Katz Saikrishnan Ramachandran March 13 2009 13,
Goals
The Objective
To decide if there is an agreement to initiate a project
Project Goals
Detect the presence of misaligned teeth i.e. outliers in sprockets Measure in line the location and the extent of deformation for each outlier relative to its neighbors while the sprocket rotates
Experimental setup
Optimet laser sensor
Sprocket
Aerotech rotary stage
Suggested Method
Measure the distance to tip of the teeth of a rotating sprocket using a non-contact single-point laser range finder Each tooth is clearly observed as a bar (several points) Outliers are observed as peaks
Evaluation of the method

Apriori Rotating speed (sprocket) Tooth width (sprocket) Diameter (sprocket) Data collection frequency (Optimet) Number of points collected/revolution Number of points per tooth/revolution w = 300 rpm L = 5 mm D = 175 mm fop = 3000 Hz N = 3000/5 = 600 n = NL/ (D) = 3000 /(175 ) = ~5.45 /(175) 5 45 = 5-6 points f = 5Hz
Conclusion: Possible to be tested !!!

Stationary Experiment
Measurement of the profile of two sets of teeth for a stationary sprocket and linearly moving Optimet sensor: one containing all normal teeth (blue) and the second containing outliers (red)
Observation : 3 outliers detected whose deformations are ~0.35 mm ~0 35 mm, 1mm and -0.21mm respectively 0 21mm

360 d degree rotary scan profile of th ti of th sprocket at a slow speed t fil f the tip f the k t t l d gives the outlier tooth location (w.r.t. starting point) and the deformation Outliers are observed at teeth 5, 6 and 7 (w.r.t. start tooth) and the measured displacements are -0.21mm; 0 98mm and 0 35mm respectively 0 21mm; 0.98mm 0.35mm

To check the performance of the method, the number of collected points per tooth is down sampled from thousands to 5 The outlier displacement values are calculated in each of the two cases Results are tabulated below Displacement Outlier 1 Outlier 2 Outlier 3 Thousands of points - 0.2088 mm 0.9899 mm 0.3585 0 3585 mm 5 random points - 0.2107 mm 0.9902 mm 0.3622 0 3622 mm
Note: negative deformation signifies inward deformation g g
Conclusion
1. The proposed method is an accurate way for the detection of outliers in sprockets 2. It is also accurate in estimating the deformation/displacement of outliers 3. Can be applied in an industrial application with a rotating sprocket
Valve Seat Gap Inspection

March13th, 2009
Project Goal
Goals:
To develop a methodology to measure the small gap between valve seat and cylinder head y
Expected Deliverables:
I li measurement technique that can enable high speed automatic In-line tt h i th t bl hi h d t ti inspection
Work done: Proof on concept using a laser probe with a motorized periscope
Infrared Valve Seat Gap p Detection System

Dan Simon Quality Network Planned Maintenance dan.1.simon@gm.com
Infrared Study of poorly seated exhaust Valve Seats
Seat gaps set @ .001, .003, g p , , 006 & .010 inches in L-6 aluminum head.
L-6 head warmed up w/ quartz lamp
Dan Simon (313) 324-5353 [ dan.simon@uaw-gm.org ]
Exhaust Valve Seat w/ .010in gap @ 113F

48.0C 48
47
46.0C
L6 Engine Head w/ .006 in gap in exhaust insert seat @ 118F Head Temperature
48.0C 48
47
46.0C
Exhaust Valve seat w/ .003 in. gap @ 120F
49.9C
49
48
47
46.0C
Exhaust Valve Seat w/ .001 in. gap @ 114 F
47.0C 47
46
45 45.0C
Exhaust Valve Seat w/ .001 in Gap @ 114 F (X 2 magnified)

47.0C 47
46
45 45.0C

Quarterly Review 031609

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Quarterly Review 031609

Caricato da

Copyright:

Formati disponibili

NSF Engineering Research Center (ERC) for

Reconfigurable Manufacturing Systems (RMS)

ERC Big Three Quarterly Review Meeting

The University of Michigan, College of Engineering

Project review History

Engineering Research Center for g g y Reconfigurable Manufacturing Systems

Reducing Unscheduled Downtime Through Automated Eventbased Control

QRM, March 16th, 2009 Indicates New Result

The University of Michigan, Ann Arbor

Project History C Current F t Focus

The Basic Idea:

This is the opportunity for fault prediction

The Basic Idea:

E ti ti Estimating MTTF: Observing failures and adjusting time to failure MTTF

Reliability Studies: Conducting independent statistical studies to model

Failure Prediction through Data Corelation

Project Objectives and Approach

Mid-Term: Predict and reduce unscheduled downtime

Approach: Closing the Loop at Global Engine

Optimize Maintenance Scheduling

OP300_1 OP110_2 OP60_3 OP160_3 OP280_3

Engineers Machine Adjustment

BENEFITS Reduced Unscheduled Downtime Reduced Scrap

CNCs Production Machines

Project History C Current F t Focus

Best practices for improving maintenance management and data q quality y

Student internships at Global Engine for Summers 06, 07 and 08

Top 10 Pareto Chart

Screen shot of viewing the top ten anomalies

Fault Total Duration

Operations with Top Fault Count

OP300_1 OP110_2 OP60_3 OP160_3 OP280_3

Cumu ulative Perce entage

Cumu ulative Perce entage

Project History C Current F t Focus

Maintenance Under Low Production Conditions

Reducing Redundant Maintenance

Failure Prediction Through Event Data:

BOLT FEEDER NOT WORKING @ 6:05 @ 15:39

@ 23:46 - Feed & Torque Station @23:46

Failure Prediction Through Event Data:

Faults on System Incur cost due to lost units, slowing production.

Alternative Schd. Monthly Downtime

Maintenance Under Lower Production Conditions:

In todays world factories are not operating at capacity

However maintenance practices are often designed for atcapacity operation

Solution: Implement part-count or event based maintenance part count event-based

Maintenance Under Lower Production Conditions: Low Load Maintenance

High Fault Rate tracks System Load

Reducing Redundant Maintenance (1)

Reducing Redundant Maintenance (2)

Reducing Redundant Maintenance (3)

Other Ongoing Efforts of Interest:

x x xx x x x x xxxxxxx xxx xx x x x xx xx xx xxx xx

Facilitates correlation analysis across many variables i bl

Other Ongoing Efforts of Interest:

Project History C Current F t Focus

Data Quality best practices improvements are starting to pay off

Current focus areas reflect economic environment

Need to determine if there is reusability

Examples of Data Quality Issues Encountered

Number of Scrap S Codes s

Samples of Recommendations for Improving Data Quality

System Wide Data Unification