Holonic planning in a high-mix, low-volume manufacturing environment
by R. Michael Mahoney, President
Manufacturing Decision Analysis, LLC
The world is facing an inescapable evolution toward high-mix, low-volume manufacturing. Customers are demanding more choices and those that fail to effectively respond to this competitive reality will cease to exist. Effective and efficient demand and execution planning are of paramount importance when the goal is to attain competitive advantage in today's competitive global marketplace. Intuitive decision-making processes are wholly inadequate and create temporal delays between often suboptimal decisions and their ramifications.
Holonic planning components of autonomous modules and their distributed control offer a powerful approach to the high-mix, low-volume demand and execution planning problem. In 1967 the word 'holon' was proposed by the Hungarian author and philosopher Arthur Koestler in his landmark book "The Ghost in the Machine." The term 'Janus Effect' described a fundamental property of sub-wholes as assertive or self-contained wholes while simultaneously dependent integrative parts within a more expansive hierarchy. Holon is a combination of the Greek word 'holos' meaning whole with the suffix 'on' suggesting a part or particle(e.g., electron). A holarchy defines a set of self-regulating holons as having three functions:
- Autonomous wholes in supraordination to their parts
- Dependent parts in subordination to higher level controls
- In coordination with their local environment
Although manufacturing complexity is increasing due to growth in product depth and breadth driven by discriminating customers, manufacturing systems are increasingly exposed to changes and disturbances. Intentional changes occur as the result of new product introductions(NPI), the move toward customized products, and the introduction of higher performance equipment and technology. Perturbations such as machine breakdowns, late raw material deliveries, stockouts, and low yields are non-intentional. Changes and disturbances are a fact-of-life and are sources of variability that a manufacturing planning and control system must cope with. Present day manufacturing planning and control systems such as Manufacturing Resource Planning(MRPII) and Enterprise Resource Planning(ERP) only work well under a strict set of conditions. Performance drops off significantly when these conditions are not met(e.g., NPI, order expediting, stockouts, machine breakdowns, etc.). Holonic planning is a flexible planning system that minimizes the need to adapt by maximizing the range of 'normal' situations.
The planning holarchy is comprised of three basic planning holons:
- Capacity holon: Includes the forecasting function of demand management. Develops the strategic annual capacity resource requirements plan(Lead, Lag, Tracking) necessary to meet the needs of the marketplace and demand response strategy[Level(L), Chase(Ch), Compromise(C)].
- Operations holon: Models the financial relationships among Capacity Utilization(C), Responsiveness(R), sequence-independent Setup Time(S), Disruptions(D), Quality loss(Q), Work-In-Progress(WIP), and relates them to the financial Loss(L) incurred over the effective WIP operating range(lot size 1�N).
- Scheduler holon: Optimizes throughput for a given product mix using the Multiple Constraint Synchronization(MCS) algorithm, uses a machine allocation heuristic, and produces products(jobs) in Shortest processing Time(SPT)sequence to minimize mean flow time and maximize the proportion of jobs delivered on-time.
The forecasting function of the capacity holon should use a simulation-based forecasting tool as opposed to a traditional stochastic forecasting tool. Stochastic forecasting tools assume smooth and continuous demand, and a normal(Gaussian bell curve) error(forecast-actual) distribution. For high-mix manufacturers, product mix can be as high as 600 unique products or greater and order quantities can vary from zero to over 1,000 over a particular defined time horizon. Stochastic forecasting assumptions will be violated.
Changes in the structure or infrastructure of a manufacturing firm will be required to augment or reduce capacity when planning annual capacity resource requirements. The lead strategy is a proactive strategy and is used when structural changes(e.g., buildings, machines, etc.) are required. The tracking strategy is a reactive strategy most often employed when infrastructural changes(e.g., additional temporary workers, increased cosourcing) are required. Incremental changes in capacity are greater for the lead strategy than the tracking strategy. A failure to strategically respond to anticipated or demonstrated increases in demand in excess of the current available capacity level will damage the delivery performance of any manufacturer.
The operations holon model shows that WIP(lot size) is the strategic driver for attaining a particular level of performance in quality, cost, delivery, responsiveness and flexibility to respond to disruptions. Lot sizing, sequencing, and allocation decisions will have a significant effect on the overall financial position of a high-mix manufacturer. As production lot sizes(WIP) are increased or decreased, there is a financial loss or benefit associated with capacity constraint utilization, quality, flexibility, and responsiveness performance. Increasing lot size(WIP) will decrease quality(e.g., batch errors), responsiveness, and delivery performance, while increasing capacity utilization. It is important to note that batch error quality problems are mitigated under the condition of increasing lot size when the transfer lot size is significantly less than the production lot sizequality problems will be detected sooner. The minimum of the total loss function is the Pareto optimum, and the degree of flexibility to increase or decrease lot sizes(WIP) about this point is controlled by a high-mix manufacturer's sequence-independent setup times.
The scheduling holon is responsible for deriving the lot size, Multiple Constraint Synchronization(MCS) sequence, and machine allocation heuristic in alignment with the resource and operations holons. Using a spreadsheet, each product is placed in SPT sequence at the constraint process step where its value-adding work content time is the greatest. SPT ensures that the mean number of jobs and mean flow time are minimal for a particular defined lot size. It is important to note that SPT is superior for reducing the proportion of late deliveries. Responsiveness of the system is based on the time to negotiate at least one unit of the entire portfolio of products ordered.
The demand response capacity strategies and the effects each strategy has on operations execution are important to understand. The forecast and production rate should be expressed as the integration(i.e., cumulative sum) of the time-series forecast. There is a significant increase in sensitivity to small changes that are difficult to detect using time-series representations. The level demand response strategy responds to forecast shipment fluctuations by using buffer inventory. The chase demand response strategy will minimize the inventory requirement and absorb forecast shipment fluctuations by changing capacity. The compromise demand response strategy represents a trade-off between the high inventory cost associated with the level strategy and the excess capacity associated with the chase strategy. The compromise demand response strategy is a postponement strategy. The initial production rate(C1) will be less than the subsequent production rate(C2). As incoming orders increase certainty about the future in the near term, the "postponed" production rate can be set more accurately. In this manner the response strategy is build-to-order rather than build-to-forecast.
The scheduler holon derives the lot size requirement and sequence required to align the executable production plan with the strategic capacity requirements and operations constraints of the company. The system least common denominator(SLCD) is calculated based on the capacity requirements necessary to cost-effectively respond to customer requirements. The MCS algorithm develops the sequence to maximize throughput given defined lot sizes. The lot size quantity(Q) necessary to amortize the global median system setup time(S) is calculated to meet the needs of the marketplace while explicitly considering capacity constraints of the manufacturing system. This quantity is multiplied by the global median work content time to derive the system least common denominator. The lot size is the ceiling[INTEGER(X+0.5)] of the SLCD divided by a particular product under consideration work content time. The sequencing of lots across the constraints is based on the MCS algorithm:
- Determine the lot with the maximum processing time at the most downstream constraint. Place this job in the first available position of the sequence.
- Remove this lot and it's associated constraint from consideration and return to step 1. If lots or constraints = 0 then stop.
Allocation of lots to parallel machines is accomplished by successively scheduling the longest remaining lot(based on value-adding work content time) to the machine where it will be completed the earliest then perform SPT sequencing of the lots associated with each machine. The resulting allocations will result in schedules that are almost always nearly optimal. This heuristic is used due to the fact that there is no known solution to the parallel machine problem short of explicitly evaluating each and every possibility. The parallel machine problem is intractable. The resulting schedule reconciles many conflicting objectives:
- Minimize cost
- Minimize mean cycle time
- Maximize throughput
- Inventory at acceptable levels
- Constraint resources well utilized
- Meet due dates
- Improve responsiveness
- Improve quality
- Robust to disruptions
The critical role of job release control is addressed through the use of generic Kanban. Forward scheduling as opposed to backward scheduling operations is a competitive imperative. Backward scheduling approaches are used in just-in-time/demand flow implementations and will damage the competitiveness of any high-mix manufacturer. Backward scheduling operations will randomize jobs at the gating process step and decrease available capacity. The output from the planning holon should be considered more as a temporal transitory state schedule than as a fixed schedule. Scheduling is performed concurrently to operations to ensure opportunities for optimization are not neglected. Thus, there is a multilevel reaction to disturbances.
Holonic planning facilitates effective judgement to ensure fast system response time to disruptions. There is heterarchical dependence among the basic planning holons rather than hierarchical control. The planning holon under heterarchical control is the executive(a.k.a. staff) holon responsible for the hierarchical control of the manufacturing planning and control system(i.e., master planning, MRP, etc.). It is imperative that the planning holon provides for feedback from the manufacturing planning and control system to allow timely and effective response to disruptions. The planning holon is a filter through which all incoming jobs must pass. This will facilitate the ability to change strategy and tactics as the situation changes and significantly improve predictability.
Manufacturing is not simply a chain of events. Supply chain assumptions are not robust to disruptions and constrain the applicability of holonic architectures. Supply chain reasoning severely limits the evolution of manufacturing planning and control systems to follow unforeseen trends.
About the Author
R. Michael Mahoney is a leading thinker on high-mix strategy, and the author of the college text High-Mix, Low-Volume Manufacturing, published through Hewlett-Packard Press by Prentice-Hall �1997. He has published numerous articles in the trade press on subjects relating to product design, process design, test, and total quality control.
Mike is President of Manufacturing Decision Analysis, LLC�A Consultancy and Seminar Company; www.HMLV.com. He has over 20 years of experience working in high-mix, low-volume manufacturing environments at Hewlett-Packard Company with broad experience from the customer site through product development to the factory floor.
He is listed in the International Who's Who of Professionals. He is a member of the Association for Manufacturing Excellence (AME), The American Society for Quality (ASQ), The Institute for Operations Research and the Management Sciences (INFORMS), The Institute of Industrial Engineers (IIE), The Institute of Electrical and Electronic Engineers (IEEE), and The American Management Association (AMA).