PRODUCTION SYSTEM (TPS /MRP) -Controlling the flow of resources by replacing only what called upon when demand is high helps to decrease
A. SIMPLE LINEAR REGRESSION MODEL : Yi = b0 + b1X1
- Method of least squares minimizes Error Sum of Squares has been consumed idle time -Resource furthest downstream (closest to market) is -Production volume: Assign some workers to processes (ESS) paced by market demand creating other products and have the remaining workers - R2 statistic indicates how well an estimated regression function fits the data 0R21 -It relays the demand information to the next station handle multiple machines simultaneously -Measures the proportion of the total variation in Y around its upstream and ensures that upstream resource is also -Product volume: Use temporary workers to help out. mean that is accounted for by the estimated regression eqn paced by demand These workers are less skilled and can only handle very TSS = ESS + RSS R2=RSS/TSS= 1- (ESS/TSS) External demand is transferred step by step through specific tasks -Standard error measures the scatter around the estimate the process
regression line oPush vs Pull System n6. STANDARDIZATION OF WORK & REDUCTION OF B. MUTLIPLE REGRESSION ANALYSIS: -Push system: has no limit imposed on downstream VARIABILITY Yi = R2 + b1X1i + b2X2i+ bkXki work-in-process inventory, whereas pull system has -In the presence of variability, need either buffer (violate Application: -Kanban system, inventory becomes a managerial 0 inventory) or suffer occasional losses in throughput Models with ONE IV decision variable (max inventory is controlled via # of (should be JIT) Suppose we have 3 simple linear regression: Kanban cards in the process) -Variability occurs because of quality defects or Yi = b0 + b1Xi Yi = b0 + b2X2i Yi = b0 + b3X3i -Push system: inventory can happen differences in activity times for the same or for different R 2 Variables Adjusted S oPull System - Kanban operators R 2 in the -Upstream replenishes what demand has withdrawn -Use quartile analysis (top vs bottom) to identify the Model from the downstream presence of large differences across operators and to X1 0.870 0.855 10.3 -Kanban= card attached to an empty container (work estimate the improvement potential X 2 0.759 0.731 14.0 authorization forms)
X 3 0.793 0.770 13.0 -Use of Kanban cards between all resources in the n7. HUMAN RESOURCE PRACTICES Model using X1 accounts for 87% of the variation in Y, process provides an effective and easy-to-implement 1. Training workers as a source of flexibility mechanism for tying the demand of the process 2. Quality circles bring workers together to jointly solve leaving 13% unaccounted for Single variable use R2 (downstream) with the production of the resources production problems and to continuously improve the (upstream) [match supply with demand] process standardize production process + PROCESS OF Models with 2 IV Suppose we have 3 simple linear regression: Can never be more inventory between 2 resources that IMPROVEMENT Yi = b0 + b1Xi + b2X2i Yi = b0 + b1Xi + b3X3i 3. Fishbone diagrams (ishikawa or cause-effect) has been authorized by Kanban cards limit the graphically represent variables that are causally related inventory to the number of containers R 2 Variables Adjusted S to a specific outcome oPull System Make-to-Order 2 R in the -Resources only operate after having received an explicit 4. Five Whys to continually question whether a potential Model cause is truly the root cause or is merely a symptom of a customer order X1 0.870 0.855 10.299 deeper problem -Each flow unit is explicitly assigned to 1 specific X1 & X2 0.939 0.924 7.471
customer order n1. SEVEN Sources of Waste X1 & X3 0.877 0.847 10.609 n8. LEAN TRANSFORMATION - 3 Aspects -Flow process on a FIFO basis 1. Overproduction: Producing too much, too soon Model using X & X accounts for 92.4% of the 1 2 1. Operating System: emphasis on flow, matching supply Companies usually use both pull systems 2. Waiting: 2 types of waiting variation in Y, leaving 7.6% unaccounted for with demand, and a close eye on the variability of the Kanban used for products or parts that are -Resource waiting for flow units idle time at resource Multiple variables use ADJUSTED R2 process 1. Processed in high volume and limited variety -Flow units wait for resources to become available **As additional IV are added to a model, the R2 can only 2. Requited with a short lead time so it makes economic 2. Management Infrastructure: Performance 3. Transport: Internal transport carrying around halfmeasurement should be tracked over time and be made increase, but adjusted R2 can increase or decrease sense to have a limited number of them preproduced finished products, wheeling patients etc - R2 can be artificially inflated by adding any IV 3. Costs and efforts related to storing the components are transparent throughout the organization. Includes the **Multicollinearity (variables highly correlated) should be Physical layout should reflect the process flow to development of operator skills and the establishment of low avoided minimize the distances flow units must travel through a formal problem-solving processes Make-to-Order used when **When fitting polynomial models, care must be taken to avoid process 3. Mindsets and Behaviors of Employees Involved: 1. Products or parts are processed in low volume and overfitting 4. Overprocessing: Spending more time on a flow unit Central to success of a lean information high variety
than necessary
2. Customers are willing to wait for their order nFORECAST 5. Inventory: Accumulation of Inventory, which is nonnTEXTBOOK QN 3. It is expensive or difficult to store the flow units 1. Forecasting Basics value-adding, & hides other problems (e.g. defects) in the 1.
-Forecasts are always wrong, should include expected process
n3. QUALITY MANAGEMENT Reduce defects value and measure of error - Tends to lead to long information turnaround times
1. Fool-proofing (Poka-Yoke) -Aggregate forecasts are more accurate 6. Rework: Rework increases variability & consumes -Poka-yoke is to fool-proof assembly operations to make A. What would happen to the inventory in the process if -LT forecasts are less accurate than ST forecasts (forecast capacity from resources the process were operated as a push system? it impossible for mistakes to happen; 1 way of horizon is important) Do it right the first time, or detect defect early In a push system, inventory would accumulate in the assembling components -Common sense compatibility 7. Motion: Many ways to perform a task system as the fast (high capacity) resource is upstream in 2. Forecasting Requirements 2. Detect-Stop-Alert (Jidoka) Only one right way to perform by fool-proofing the process. This would be non-value-adding, as it would -Accurate with small errors -To stop process immediately whenever a defect is Ultimately to reduce waste so as to increase output (flow have no positive impact on the overall flow rate. detected and to alert line supervisor -Unbiased, so they do not always under or over estimate rate) and revenues in a capacity constrained process B. Assuming you would have to operate as a push system, demand -Attempts to discover and isolate problem ASAP
how would you resequence the three activities? Andon card, well-known form of Jidoka -Responsive to changes in demand n2. JIT: MATCHING SUPPLY WITH DEMAND -A cord running adjacent to assembly lines that enables Since the sequence of the assembly tasks does not matter, -Not affected by outliers 3 Steps toward achieving a JIT process we could simply move the slowest step first. This would -In time for its purpose workers to stop production if they detect a defect 1. Achieve a One-Unit-at-a-Time Flow ensure that we are not moving flow units into the system -Cost-effective Allows for build-in-quality within the process vs -Mixed-Model Production: Product variants are mixed faster than the bottleneck pace. relying on final inspection step for defects -Easy to understand together on the assembly line C. How would you implement a pull system? -Jidoka benefits from 0 inventory principle 3. Influencing Factors -Creates steady workloads -Size of buffer between 2 resources affects time between A pull system could be implemented using Kanban cards -Time covered in the future -Creates steady demand for parts when the problem occurred at the upstream resource and between the stations. This way, the slow last step could -Availability of historical data -Made possible by aggressive setup time reduction the time it is detected at the downstream resources [ITAT pace the faster steps upstream. We should also ensure -Relevance of historical data to the future -Piece-by-Piece Transfer: transfer batches of one that we are not producing faster than the overall demand -Type of product = information turn-around time] -Made possible by process layouts with limited distance 2. How to improve this process? -Smaller buffer = shorter time -Variability of demand between steps (U layout)
-Accuracy needed and cost of errors -Avoids unnecessary transports and reduces the need to
n4. EXPOSING PROBLEMS THRU INVENTORY -Amount of money and time available for the forecast form transport batches
REDUCTION 4. Approaches to Forecasting 2 resources that are close to each other in the process Redesign the lay-out of the process, holding the current Forecasting methods Qualitative vs Quantitative Buffer argument vs Toyota argument flow diagrams should be co-located in physical space sequence constant. By creating a U-shaped line, it is -Buffer argument: Increase inventory (Projective vs Causal) 2. Produce at the Rate of Customer Demand possible to have step 6 and step 1 be staffed by the same A. Qualitative/Judgmental Forecasting -Toyota argument: Decrease inventory -So as to reduce finished goods inventory person, who would operate on the 3 min cycle that is -Kanban system allows managers to gradually reduce -Personal insight from an expert -Customer demand = Production rate (Flow rate) using imposed by the bottleneck step (step 3). This would cut -Panel consensus inventory to expose quality problems takt time Step 1: Reduce inventory; but has ST losses in throughput the cost of direct labor/ it would increase the average -Market surveys 3. Implement a PULL SYSTEM using Kanban or make-toStep 2: With lowered inventory level, problems are easy labor utilization. -Historical analogy order production
to identify and to be solve -Delphi method oPush System
Step 3: When problems are solved, this moves the B. Quantitative Forecasting -Resources are provided to the consumer based on
process to a better frontier -Time-series Approach: series of observations taken at forecasts or schedules
DEMAND FORECASTING regular time intervals -Parts completed to schedule w/o any downstream n5. FLEXIBILITY WORKERS WITH VARIOUS SKILLS nRegression Analysis is used to estimate a function f(x) -3 common patterns: Constant, Trend, Seasonality considerations -TPS attempts to create process with sufficient flexibility that describes the relationship between a continuous DV -Upstream resource continues to produce as long as it to meet demand fluctuations and one or more IV
has work (input) -Have multitiered workforce consisting of highly skilled Y = f(X1, X2, X3, Xn) +
oPull System full-time employees AND a pool of temporary workers f(x): systematic variation in the relationship
(who do not need such a broad skill base) that can be : unsystematic variation (random error) in the rship
Zero non-value added activities (muda) 0 defects, 0 breakdowns 0 inventory, 0 set-up Production flow Quality methods to synchronized with reduce defects demand (JIT) -Fool-proofing (poka-One-unit-at-a-time yoke) & Visual feedback flow -Detect-stop-alert 1. Mixed model pdn (Jidoka) 2. Piece-by-piece trf -Defects at machines -Match production (original Jidoka) demand with Takt time -Defects in assembly -Pull instead of push (Andon cord) 1. Kanban -Build-in-Quality 2. Make-to-order (Tsukurikomi)
- Adjustment of capacity to Flexibility meet Takt time - Multitask assignment - Reduction of variability Standardization of - Quartile analysis Work - Standing operating procedures - Quality circles Worker - Fishbone diagrams Involvement - Skill Development/ X-training
MEASURES OF FORECAST ERROR
-Result in changes of the parameters of the underlying statistical distribution of the process ei = Demand - Forecast -Due to differences in variance or mean
-Need to assign chances so as to prevent them 1. Mean forecast error (MFE)= nImpact of Defects measures bias 1. Lose capacity of the process (can become the
bottleneck, or worse, defect at bottleneck) 2. Mean absolute deviation (MAD)= 2. Internal failure costs: related to testing, reworking show on average how far a forecast capacity constraints is away from the actual value 3. External failure costs: lost reputation
nWhat to do about Defects? 3. Mean squared error (MSE)= 1. Detect early is better than detecting late penalizes the outlier -capacity can increase and costs can reduce
CAUSAL FORECASTING: Look for a cause or relationship -# of defects can be reduced -cause can be identified more easily -Use Regression Analysis 2. Prevention better than inspection PROJECTIVE FORECASTING -Inspection adds non-value added costs and capital -Prediction based on previously observed values -Does not remove costs of scrap or rework -Detect patters -Is never perfect (defects slip through) -ST demand prediction -3 types: SA, MA, ES [use forecast error to measure which -Removes responsibility -Can become a bottleneck method is better] nREWORK 1. Simple Average Process yield= Flow rates of units processed successfully 2. Moving Average (MA)
Flow rate -Used when demand has no observable trend or = 1 (Flow rate of defects/ Flow rate) seasonality -Rework changes the utilization profile of the process -Systematic component of demand = constant level -The level in period t is the average demand over the last =A resource has additional work flowing to it, which increases utilization N period (the N-period moving average) -Current forecast for all future period is the same and is -If rework involves only nonbottleneck a large amount based on the current and is based on the current estimate of idle time (negligible effect on overall process capacity, but still have cost implications) of the level -If rework involves bottleneck or causes a new bottleneck -Large N Stability up capacity invested in rework at bottleneck is lost from -Small N Responsiveness up the perspective of overall process 3. Simple Exponential Smoothing (ES) nEliminating Flow Units from the Process - Used when demand has no observable trend or To get 100 good (demand) units 100/process yield seasonality Process yield = y1 x y2 x x ym -Systematic component of demand = constant level -New forecast = weighted average of last demand and last Where m is the number of resources in the sequence Produce more units to account for defects forecast
-Large N Stability up
-Small N Responsiveness up (not necessarily more
accurate due to random effects of noise) *New Forecast = x Latest demand/Actual sales + (1- ) nControl Charts to visualize variation x chart: Y-axis corresponds to the mean of each sample x Previous Forecast * is the smoothing constant (usually between 0.1-0.2) R chart: Y-axis corresponds to the range of each samply Let Yt = actual sales in week t R= max (X1, X2, Xn) - min (X1, X2, Xn)
Ft = forecasted sales in week t
F1 = Y1 #1: Compute mean and range for each period For other weeks, Ft+1 = Yt + (1- ) Ft #2: Compute average mean (X) and average range (R)
Week Sales Week Sales R & X are CENTRE LINE for respective graphe
1 110 3 125 = 0.1 #3: Find UCL & LCL, set at x% confidence 2 115 4 120
UCL for X = X + a2 x R LCL for X = X - a2 x R F1 = Y1
= 110 UCL for R = d3 x R F2 = 0.1Y1 + 0.9F1 = 0.1(110) + 0.9(110) = 110 LCL for R = d4 x R F3 = 0.1Y2 + 0.9F2 = 0.1(115) + 0.9(110) = 110.5
C : more capable the process less defects, but more p Similarities MA & ES ex - Stationary process nPareto Analysis: helps to decide which assignable - One parameter (N or ) causes to eliminate first - Lag behind trend 80/20 rule: 80% of defects are the result of only 20% of Differences defect types -ES: All past data TB QN -MA: Last N periods (outliers washed out) We define four types of flow units:
- type a: breaks at #1 (20% of total) -ES: Less data (last forecast x demand) - type b: breaks at #2 (80%*10%=8%, has to pass #1, but -MA: Last N data then breaks at #2)
- type c: breaks at #3 (80%*90%*10%=7.2%) QUALITY MANAGEMENT - type d: does not break at all, i.e. does not require n2 Types of Quality rework (80%*90%*90%=64.8%) 1. Design quality is embedded in the design specifications Resource Minutes of the product of service available 2. Conformance quality depends on the production and Type a Type b Type c Type d Workload delivery processes (operations) 1 60 0.2*6 0.08*6 0.072*6 0.648*6 6 Non-conformance quality = Defects caused by variation 2 60 0 0.08*5 0.072*5 0.648*5 4 nVariation 3 60 0 0 0.072*4 0.648*4 2.88 -Needs to be controlled Rework 60 0.2*15 0.08*10 0.072*5 0 4.16 -Some variation is acceptable (a) If flow units arrive at a rate of 5 units per hour, we -Process not capable when there are too many defects have 4 units per hour arriving at #2 (the 1. Common Causes other 1 unit will on average break at #1). Thus, the -Constant variation reflecting in pure randomness in the utilization there is: flow process rate/capacity=4/12 -Cannot predict exact outcome for the randomness in (b) The bottleneck is #1 every single flow unit (c) We can process up to 10 units per hour, at which 2. Assignable Causes point the utilization of #1 hits 100%. Note