dortmund university of technology Pushing the Limits of Lean Thinking by Simulation and Data Analytics Univ.-Prof. Dr.-Ing. Jochen Deuse Stuttgart, 16 June 2015 1 Institute of Production Systems Research Areas Time and Motion Studies Digital Manufacturing Manufacturing Data Analytics Human-Robot- Collaboration Work System Design Factory Physics 2
Basic rules of Lean Thinking (Toyota-DNA) Decoding the DNA of the Toyota Production System Rule 1: All work shall be highly specified as to content, sequence, timing, and outcome. Stabilise Rule 2: Every customer-supplier connection must be direct, and there must be an unambiguous yes-or-no way to send requests and receive responses. Pull Rule 3: The pathway for every product and service must be simple and direct. One-Piece-Flow Rule 4: Any improvement must be made in accordance with the scientific method, under the guidance of a teacher, at the lowest possible level in the organisation. Single-Factor Experiments [Spear & Bowen 1999] 3 Basic rules of Lean Thinking (Toyota-DNA) Decoding the DNA of the Toyota Production System Rules are formulated generically and therefore they suggest universal transferability. However, the rules must be interpreted in a domain-specific context. Context is defined by Value Added Variability (VAV). Methods have to be chosen depending on features and expressions of variability, but also principles have to be scrutinised. 4
Categories of Variability Does not contribute to added value Loss of availability Loss of quality Loss of performance should be eliminated Does not contribute to added value Probabilistic risk Common Cause Weather Environmental impacts can t be eliminated entirely Contribute to added value Customised delivery quantity Customer specific due-date Customised product specification should not be eliminated 5 Features of Value Added Variability (VAV) Features of Variability Sequence of operations Process routing Work content Target lead time Demand Measurement parameters Similarity: adapted Jaccard-coefficient Dispersion: coefficient of variation 6
Domains of Production Systems defined by VAV Variability of target lead time Machine Tools & Industrial Equipment Automotive-OEM 2013 [Daimler 2010] Toyota 1987 Ford 1913 [dapd/ronald Wittek 2013] [AP/Paul Sakuma] [Hulton Collection] Variability of work content 7 Classification of Production Systems acc. to Cynefin Framework Unordered Dynamic cause-effect relationship Complex: high VAV Experimental/ Explorative Ordered Static cause-effect relationship Complicated: high VAV Analytical/ Experimental Chaotic probe sense respond sense analyse respond Simple/ Obvious: low VAV Apply existing methods and tools act sense respond sense categorise respond [acc. to Snowden 2007] 8
Classification of Production Systems acc. to Cynefin Framework Unordered Dynamic cause-effect relationship Complex: high VAV Experimental/ Explorative Ordered Static cause-effect relationship Complicated: high VAV Analytical/ Experimental Chaotic probe sense respond sense analyse respond Simple/ Obvious: low VAV Apply existing methods and tools act sense respond sense categorise respond [acc. to Snowden 2007] 9 Simple Production Systems Use case Industrial Engineering-Training Centre: Gear Box Assembly Line Ordered, simple/ obvious System Simulated work environment, i. e. laboratory Simple and well defined tasks Simple and static cause-effect relationships Categories of Variability Non-VAV: Loss of availability Loss of quality Loss of performance VAV: None Select and apply existing methods and tools (Lean Recipes), e.g. value stream mapping, line balancing or Kanban 10
OXOX Tecnomatix Plant Simulation Worldwide User Conference 2015 Applying Pencil & Paper Methods Value Stream Mapping Line Balancing 11 Stabilise Value Streams: Eliminate Mura Non value added variability (Mura) is a major root cause of waste (Muda) Stabilisation of bottleneck Buffer against variability in supply Buffer against variability in demand Pre-Process 2 Pre-Process 1 Bottleneck Gradual stabilisation of pre-processes Start: stabilise the bottleneck [Richter & Deuse 2011] 12
SZ[min]: Ø=27, =61 Stabilise: Dynamic Value Stream Mapping Quantify Variability within Value Streams FW (03) NG: Ø= 87%; =21% VZ[min]: Ø=146, =126 PZ[min]: Ø=92, =50 SZ[min]: Ø=53, =110 NG: Ø= 67%; =13% FG (37) VZ[min]: Ø=612, =316 PZ[min]: Ø=503, =253 SZ[min]: Ø=125, =112 Sperrlager QS-Lager FW (06) VZ[min]: Ø=193, =77 PZ[min]: Ø=128, =41 SZ[min]: Ø=65, =55 NG: Ø= 54%; =13% NG: Ø= 90%; =21% FG (34) VZ[min]: Ø=413, =209 PZ[min]: Ø=341, =145 SZ[min]: Ø=73, =178 STR (07) VZ[min]: Ø=67, =223 PZ[min]: PZ = 35 min Ø=35, =17 SZ[min]: Ø=32, =222 PZ = 17 min NG: Ø= 52%; =14% V PZ = 0,49 FS (44) VZ[min]: Ø=108, =49 PZ[min]: Ø=53, =24 SZ[min]: Ø=55, =40 NG: Ø= 34%; =10% NG: Ø= 76%; =27% Ø= 106 h = 52 h LZ = 50 h 36 (PZ): Ø=5,7 h, =2,2 h 03 (PZ): Ø=1,5 h, =0,8 h 37 (PZ): Ø=8,4 h, =4,2 h 06 (PZ): Ø=2,1 h, =0,7 h LZ = 69 34 h(pz): Ø=5,7 h, =2,4 h Ø= 50 h V LZ = = 69 1,38 h 07 (PZ): Ø=0,6 h, =0,3 h 44 (PZ): Ø=0,9 h, =0,4 h Ø= 35 h Ø= 58 h = 56 h = 108 h Simulate Value Streams using Plant Simulation 13 Classification of Production Systems acc. to Cynefin Framework Unordered Dynamic cause-effect relationship Complex: high VAV Experimental/ Explorative Ordered Static cause-effect relationship Complicated: high VAV Analytical/ Experimental Chaotic probe sense respond sense analyse respond Simple/ Obvious: low VAV Apply existing methods and tools act sense respond sense categorise respond [acc. to Snowden 2007] 14
Horizontal fraises machine Washing Vertical lathe (M/C 2) Vertical lathe (M/C 1) Welding Radial drilling machine Housing assembly C Housing assembly B Housing assembly A Gantry Blade assembly 4 Blade assembly 3 Blade assembly 2 Blade assembly 1 Disassembly Horizontal fraises machine Washing Vertical lathe (M/C 1&2) Vertical lathe (M/C 1&2) Welding Radial drilling machine Housing assembly C Housing assembly B Housing assembly A Gantry Blade assembly 4 Blade assembly 3 Blade assembly 2 Blade assembly 1 Disassembly Tecnomatix Plant Simulation Worldwide User Conference 2015 Complicated Production Systems Use case: Manufacturing of steam turbine components Ordered, complicated system Cause and effect typically temporally and spatially distinct Different products with different processes and sequences of operations Make-to-Order setting, i. e. unpredictable order sequence Categories of Variability Non VAV: Loss of availability, quality or performance Environmental impacts VAV: Sequence of operations Process routings Work content Demand Lead time Analytical/ Experimental: Apply Factory Physics and conduct simulation experiments; keep True North of Lean Thinking in mind 15 One-Piece-Flow: Layout Housing Manufacturing Optimise factory layout according to Lean Thinking: The pathway for every product and service must be simple and direct. Using simulation experiments to identify capacity overload (Muri) Factory Layout and Flow of Material Simulation Experiments M/C 2 M/C 2 M/C 1 M/C 1 Simulation Results Factory Layout and Flow of Material Inventory Utilisation M/C Pool pooling of bottleneck machines disjunctive [RIF e. V.] 16
Factory Physics: Experiments on Kingman s Formula 17 Classification of Production Systems acc. to Cynefin Framework Unordered Dynamic cause-effect relationship Complex: High VAV Experimental/ Explorative Ordered Static cause-effect relationship Complicated: High VAV Analytical/ Experimental Chaotic probe sense respond sense analyse respond Simple/ Obvious: Low VAV Apply existing methods and tools act sense respond sense categorise respond [acc. to Snowden 2007] 18
Complex Production Systems Use case: Machining of engine components in a flexible manufacturing system Unordered, complex System Causalities are dynamic, but discoverable Growing number of system elements, interfaces and systeminternal relationships High level of automation Categories of Variability Non VAV: Varying system load Dynamic bottlenecks Loss of availability, quality or performance VAV: Sequence of operations Process routings Work content Demand Lead time [Felsomat 2014] Experimental/ Explorative: Manufacturing Data Analytics based on unsupervised and supervised machine learning 19 Limited applicability of Scientific Method Apply probing instead of Single-Factor Experiments Identify unknown, multivariate patterns in industrial databases Use data mining, i.e. unsupervised learning procedures [Felsomat 2014] System State: Unstable Stable [RIF e. V.] 20
Trends in Big Data Analytics Data Volume (in bytes) Zetta (10 21 ) Exa (10 18 ) Unstructured External Data High Dimensional Data Real-Time Analysis Predictive Data Analysis Peta (10 15 ) Unstructured Internal Data Tera (10 12 ) Traditional Instructions Real-Time Report Process Automation Ad hoc decision Support Operational Planning Strategic Planning Years Seconds Years Analysis Time Line [acc. to Schwab and Keil 2012] Due to long run-times, simulation experiments are not applicable in realtime and predictive analysis Supervised machine learning supports real-time analysis Learning of prediction model by using simulation data 21 Conclusion Complex High VAV Probing: unsupervised machine learning Combination of simulation and supervised machine learning Complicated High VAV Apply Factory Physics and conduct simulation experiments Keep True North of Lean Thinking in mind [Felsomat 2014] Level of IT-Support AP/Carmelo Imbesi [Kurier.at 2014] Chaotic No manageable patterns No occurrence in production systems Simple/ Obvious Low VAV Select and apply existing methods and tools 22
Thank you for your kind attention! Prof. Dr.-Ing. Jochen Deuse TU Dortmund University Institute of Production Systems jochen.deuse@ips.tu-dortmund.de www.ips.do 23