Jonathan Gaudreault

Study of the performance of multi-behaviour agents for supply chain planning

In todays industrial context, competitiveness is closely associated to supply chain performance. Coordination between business units is essential to increase this performance, in order to produce and deliver products on time to customers, at a competitive price. While planning systems usually follow a single straightforward production planning process, this paper proposes that partners adapt together their local planning process (i.e. planning behaviours) to the different situations met in the supply chain environment. Because each partner can choose different behaviour and all behaviours will have an impact on the overall performance, it is difficult to know which is preferable for each partner to increase their performance. Using agent-based technology, simulation experiments have been undertaken to verify if multi-behaviour planning agents who can change planning behaviours to adapt to their environment can increase supply chain performance. These agents have been implemented in an agent-based planning platform, using a case study illustrating a lumber supply chain. The performance analysis shows that advanced planning systems can take advantage of using multiple planning processes, because of the dynamic context of supply chains.

Parallel discrepancy-based search

Backtracking strategies based on the computation of discrepancies have proved themselves successful at solving large problems. They show really good performance when provided with a high-quality domain-specific branching heuristic variable and value ordering heuristic, which is the case for many industrial problems. We propose a novel approach PDS that allows parallelizing a strategy based on the computation of discrepancies LDS. The pool of processors visits the leaves in exactly the same order as the centralized algorithm would do. The implementation allows for a natural/intrinsic load balancing to occur filtering induced by constraint propagation would affect each processor pretty much in the same way, although there is no communication between processors. These properties make PDS a scalable algorithm that was used on a massively parallel supercomputer with thousands of cores. PDS improved the best known performance on an industrial problem.

Combined planning and scheduling in a divergent production system with co-production: A case study in the lumber industry

Many research initiatives carried out in production management consider process planning and operations scheduling as two separate and sequential functions. However, in certain contexts, the two functions must be better integrated. This is the case in divergent production systems with co-production (i.e. production of different products at the same time from a single product input) when alternative production processes are available. This paper studies such a context and focuses on the case of drying and finishing operations in a softwood lumber facility. The situation is addressed using a single model that simultaneously performs process planning and scheduling. We evaluate two alternative formulations. The first one is based on mixed integer programming (MIP) and the second on constraint programming (CP). We also propose a search procedure to improve the performance of the CP approach. Both approaches are compared with respect to their capacity to generate good solutions in short computation time.

Optimization/simulation-based framework for the evaluation of supply chain management policies in the forest product industry

This work describes a framework for the elaboration and evaluation of management policies for production and transportation supply chains in the forest product industry. The approach deals with the issue of coordination between the tactical and operational decision levels. First, we introduce LogiLab, a software system allowing to model the network and to optimize product flows in the supply chain. We than show how one can use this tactical aggregated plan to identify management policies that will guide day to day operations at the operational level. Finally, a discrete event simulation model allows assessing with more details what would be the impact of implementing these policies at the operational/execution level.

Parallel Depth-Bounded Discrepancy Search

Search strategies such as Limited Discrepancy Search (LDS) and Depth-bounded Discrepancy Search (DDS) find solutions faster than a standard Depth-First Search (DFS) when provided with good value-selection heuristics. We propose a parallelization of DDS: Parallel Depth-bounded Discrepancy Search (PDDS). This parallel search strategy has the property to visit the nodes of the search tree in the same order as the centralized version of the algorithm. The algorithm creates an intrinsic load-balancing: pruning a branch of the search tree equally affects each worker’s workload. This algorithm is based on the implicit assignment of leaves to workers which allows the workers to operate without communication during the search. We present a theoretical analysis of DDS and PDDS. We show that PDDS scales to multiple thousands of workers. We experiment on a massively parallel supercomputer to solve an industrial problem and improve over the best known solution.

Design of Multi-Behavior Agents for Supply Chain Planning: An Application to the Lumber Industry

New economic challenges and recent trends regarding globalization have forced companies of many industries, including the Canadian lumber industry, to question aspects of their organizations. Many of them have looked to reengineer their organizational processes and business practices and adopt supply chain management best practices. An aspect studied by many researchers recently is supply chain sales and operations planning, which deals with the management of client orders through the supply chain. Each partner involved must decide quantities and production dates, and allocate resources for each product needed, with respect to production capacities and transportation delays. Coordination between production partners is essential in such a context in order to deliver products on time to final clients. As perturbations occur all the time in such complex system, production centers have to react quickly to correct deviances and create new plans, while coordinating changes with partners. At the structural level, centralized approaches handle supply chain planning and coordination with difficulty, mainly because of the complexity of such problems and the challenges of sharing private information between partners. Decentralized approaches are now being considered to overcome these problems, giving different partners the responsibility to locally plan their production, using coordination schemes to insure coherent supply chain behavior. Agent-based technology provides a natural approach to model supply chain networks and describe specialized planning agents. On the other hand, decentralized approaches are generally sub-optimal. Heuristics are used by agents to coordinate and optimize their production plan in order to reach feasible global solutions. Because a local change in a plan can impact other partners, a coordination mechanism must be used to insure that every partner is informed of the change and can make their own changes if necessary. Most of the time, system designers or production planners select a planning heuristic at design time, choosing what they believe to be the best decision for their specific application. The main problem is that the heuristic may not be adapted to further perturbations or environmental conditions the planning agents will face in a production context. Usually, these local algorithms used by agents can be parameterized on several levels (such as

A periodic re-planning approach for demand-driven wood remanufacturing industry: a real-scale application

This article develops an experimental platform to select production planning policy in demand-driven wood remanufacturing industry. This industry is characterised by divergent co-production, alternative processes, a make-to-order philosophy and short order cycle times. Under such complex characteristics, the selection of an efficient production plan is a complex task. Previous work has failed to address all the industrial characteristics encountered in wood remanufacturing mills. After defining key performance indicators (KPIs) to measure the production plan efficiency, our methodology uses a periodic re-planning strategy based on a rolling horizon. Then, mixed-integer programming models are formulated leading to different planning approaches. Finally, the resulting decision framework is experimented to prescribe the best planning policy based on the selected KPI. Each production planning is characterised by its planning approach and factors related to the re-planning interval and the planning horizon length. Simulations are conducted using multiple best subset selections combined with an experimental design approach. Using industrial data from a wood remanufacturing mill in Eastern Canada, results indicate that the manufacturing mill should use a planning approach that minimises cost, while utilising the full system capacity. Results also quantify the benefit of using lower re-planning intervals and higher planning horizons.

Supply Chain Coordination Using an Adaptive Distributed Search Strategy

A tree search strategy is said to be adaptive when it dynamically identifies which areas of the tree are likely to contain good solutions, using information that is gathered during the search process. This study shows how an adaptive approach can be used to enhance the efficiency of the coordination process of an industrial supply chain. The result is a new adaptive method (called the adaptive discrepancy search), intended for search in nonbinary trees, and that is exploitable in a distributed optimization context. For the industrial case studied (a supply chain in the forest products industry), this allowed reducing nearly half the time needed to obtain the best solution in comparison with a standard nonadaptive method. The method has also been evaluated for use with synthesized problems in order to validate the results that are obtained and to illustrate different properties of the algorithm.

Evaluating order acceptance policies for divergent production systems with co-production

The impacts of using different order acceptance policies in manufacturing sectors are usually well known and documented in the literature. However, for industries facing divergent processes with co-production (i.e. several products produced at the same time from a common raw material), the evaluation, comparison and selection of policies are not trivial tasks. This paper proposes a framework to enable this evaluation. Using a simulation model that integrates a custom-built ERP, we compare and evaluate different order acceptance policies in various market conditions. Experiments are carried out using a case from the forest products industry. Results illustrate how and when different market conditions related to divergent/co-production industries may call for available-to-promise (ATP), capable-to-promise (CTP), and other known strategies. Especially, we show that advanced order acceptance policies like CTP may generate a better income for certain types of market and, conversely to typical manufacturing industries, ATP performs better than other strategies for a specific demand patterns.

Human-machine interaction for real-time linear optimization

Mixed-Initiative-Systems (MIS) are hybrid decision-making systems in which human and machine collaborate in order to produce a solution. This paper described an MIS system adapted to business optimization problems. These problems can be solved in less than an hour as they show a linear structure. However, this delay is unacceptable for iterative and interactive decision-making contexts where users need to provide their input. Therefore, we propose a system providing the decision-makers with a convex hull of optimal solutions minimizing/maximizing the variables of interest. The users can interactively modify the value of a variable and the system is able to recompute a new optimal solution in a few milliseconds. Four real-time reoptimization methods are described and evaluated.