420 likes | 560 Views
SPC for Services: Timeliness and Correctness Monitoring. Russell Barton Department of Supply Chain and Information Systems The Pennsylvania State University
E N D
SPC for Services: Timeliness and Correctness Monitoring Russell Barton Department of Supply Chain and Information Systems The Pennsylvania State University Acknowledgments: John McCool. Jun Shu, Earnest Foster, Jeff Tew, Lynn Truss, Smeal College Center for Supply Chain Research National Science Foundation
Overview • What do we mean by service quality? • Process Execution Monitoring: “SPC for Services” • Optimization versus monitoring views • Process execution monitoring: supply chain timeliness and correctness • The work to be done 2
Supply Chain: a Service Process Customers Retailers Warehouse/Dist Manufacturer Suppliers Suppliers’ Suppliers Source: www.dallasfed.org/research/swe/2005/swe0502b.html 3
Another Service Process: Mortgage Application LE C TS CR LE C RC TS LD CR TS C LE Customers Customer Reps References/Credit Title Search Loan Design Loan Execution 4
A (narrow) Service Process View • Transactions moving through process steps: • a mortgage application moving through credit check, title search, loan design • a business order moving through order assembly, packing, loading, shipping, unloading, unpacking • Two key characteristics: • how much time in each step • correctness of sequence of steps
Service Quality • Timeliness of Service Processes • Entity or transaction time in a particular location (state) • Entity or transaction time between locations or states • Correctness of Service Processes • Entity processed through a correct sequence of steps or locations (states) • There may be more than one correct sequence • The sequence often depends on the kind and/or ID of the entity 6
Service Quality • Timeliness and Correctness characterize many types of service operations: • Processing a mortgage • Delivering a package • Cleaning an office building • Providing emergency room treatment • Providing an educational certificate or degree • Providing airline service • A supply chain operation
Process Execution Monitoring: SPC for Services • Idea: apply SPC and process capability methods to timeliness and correctness measures from service process execution data • For semi-automated processes this is a special kind of Workflow Monitoring • For the remainder of this presentation, we will focus specifically on supply chain processes, but the approach can be applied to any transaction processing system
Control Chart Basics Out of Control → UCL LCL Time → = a statistic (individual value, average, range, std. dev.) for a subgroup of performance data 9
Cpk = min (USL – avg, avg – LSL) = 2.5/3 3 Process Capability Basics LSL avg USL 10
SPC for Supply Chains: the Need • Need for SPC/Capability • Are your suppliers’ deliveries repeatable? • What is their process capability relative to delivery time windows? • Can you detect changes (‘out of control’) in the delivery timeliness before there is a crisis? • What stages of the delivery process cause the greatest variation in delivery time? How much might delivery time variation be reduced? • How do you tell on a daily or hourly basis which parts of your supplier chains or delivery chains need attention? 11
Contrasting Process Execution Monitoring with the usual Supply Chain Management Focus: Optimization versus Monitoring 12
Enterprise structure (distributed RFID read data) RFID simplified structure 1 001 12:00 1 001 12:01 1 001 12:02 1 002 12:02 1 003 12:02 1 001 12:03 1 001 12:00 1 001 12:01 1 001 12:02 1 002 12:02 1 003 12:02 1 001 12:03 1 001 12:00 1 001 12:01 1 001 12:02 1 002 12:02 1 003 12:02 1 001 12:03 2 001 12:00 2 001 12:01 2 001 12:02 2 002 12:02 2 003 12:02 2 001 12:03 1 001 12:00 1 001 12:01 1 001 12:02 1 002 12:02 1 003 12:02 1 001 12:03 1 001 12:00 1 001 12:01 1 001 12:02 1 002 12:02 1 003 12:02 1 001 12:03 3 001 12:00 3 001 12:01 3 001 12:02 3 002 12:02 3 003 12:02 3 001 12:03 1 001 12:00 1 001 12:01 1 001 12:02 1 002 12:02 1 003 12:02 1 001 12:03 4 001 12:00 4 001 12:01 4 001 12:02 4 002 12:02 4 003 12:02 4 001 12:03 5 001 12:00 5 001 12:01 5 001 12:02 5 002 12:02 5 003 12:02 5 001 12:03 Core of Supply Chain Execution Data: the SIT Triple • Abstract view: SIT triple • S: state (RFID reader location) • I: ID for entity (Case ID) • T: time stamp 14
Using SIT Data to Monitor Timeliness and Correctness • Sets of raw (s, i, t) data can be used to characterize ‘timeliness’ and ‘correctness’ • Use ‘echoset’ and ‘neighborhood’ concepts • To aggregate multiple reads • To determine arrival to and departure from a readable state • Infer entrance to and departure from nonreadable states • To allow calculation and characterization at different levels of aggregation 15
SIT Data • The plot shows RFID reads for 10 items at one reader location, over time.
SIT Data and Timeliness • The boxes indicate echosets of RFID reads, considered as an aggregate presence of a transaction (or item) at a particular state over a period of time
SIT Data and Timeliness Order 4 Order 3 Order 2 Order 1 • This neighborhood is a collection of four echosets (IDs from the same order in the same echoset) that have specified characteristics.
SIT Data and Timeliness Order 4 Order 3 Order 2 Order 1 • Timeliness is measured by sojourn time of an echoset or averaged over a neighborhood of echosets
SIT Data and Sequence Correctness • Correctness requires a three-dimensional view of the SIT triple • The next figure collapses multiple states onto the vertical axis, which now capture both state and id… • For these items, the correct sequence is state S1, then state S2, then state S4. • Four groups have their data in the plot, resulting in two correct sequences (S1, S2, S4) and two incorrect sequences (S1, S4) and (S1, S3, S2, S4) – can you see it?
Recall SIT Data and Timeliness Plot Order 4 Order 3 Order 2 Order 1
Monitoring Correctness • Measuring path correctness involves comparing an actual sequence of states to one or more prescribed sequences. • There are a number of algorithms for measuring such matches, coming from fields such as language processing and genome sequencing. One example is Edit Distance. • These algorithms generally rely on some form of dynamic programming, and are computationally tractable for a small number of sequence steps. 26
With these data we can plot the subgroup average sequence error: SPC for Sequence Correctness!
SPC for Supply Chains:If Straightforward, Why is there Little Use? • Difficulties: • Availability of data • Form of data • Multivariate data (different shipment modes, products, destinations) • Dependencies (multiple items in same truck) • Defining measures of timeliness and correctness at multiple scales • Inherent time lags and censoring 29
SPC for Supply Chains:Difficulties • Some Ideas: • Dependencies (multiple items in same truck) • Inherent time lags and censoring 30
Identifying Network-Based Dependencies from Group Movements and other Causes • If traveling common links is the major source of covariance in times, efficient methods are available to estimate covariances for different items sharing all or part of their routes. • Variances (and perhaps covariances) in individual links paired with topology are sufficient to estimate all path covariances. 31
xi yi Network-based Covariance • Entities traveling from 1-5 and 2-6 always share 3-4 1 5 wi 3 4 xi = s1-s5 time = wi + vi yi = s2-s6 time = wi + bi Cov(X, Y) = Var(W) 2 6 32
xi yi Network-based Covariance • More realistic: entities traveling from 1-5 and 2-6 sometimes share 3-4 1 5 wi 3 4 ai xi = s1-s5 time = wi + vi yi = s2-s6 time = ai + bi Cov(X, Y) = Cov(A, W) 2 6 33
Efficiency of Common Link Covariance Estimators • Let C1 be the usual covariance estimator based on xi and yi, and C2 be common link estimator based on ai and bi. • Then Var(C1) = Var(C2) + Var(Q+R+S) • Where Q, R, S are the usual estimators for Cov(V,A), Cov(V,B) and Cov(W,B) respectively
SPC for Supply Chains:Difficulties • Some Ideas: • Inherent time lags and censoring 35
Determining Sojourn Time at S for I Items in I sojourn Time → 36
Determining Sojourn Time at S for I Items in I Time → sojourn 37
Determining Sojourn Time at S for I Items in I Time → ’20%’ sojourn 38
Censored Data Issue:Determining Sojourn Time at a Particular State Subset S for Item Subset I Items in I Time → sojourn 39
SPC for Supply Chains: Work to be Done • Identification of technology gaps and roadblocks to implementation (data access, data cleaning, data structure) • Research on modifications to SPC and capability tools to apply to supply chain data: dependence and censoring • Develop best presentation formats (dashboards) for capability and control analyses to enable effective supply chain management 40
Questions? 41