⚡ electrical engineering

Release Date Optimization in MRP Using Clearing Functions

This paper proposes an enhanced Material Requirements Planning (MRP) approach that integrates clearing functions to replace traditional backward scheduling, thereby significantly improving performance in flow shop systems by dynamically computing planned lead times for individual production orders under varying demand and capacity constraints.

Original authors: Wolfgang Seiringer, Klaus Altendorfer, Reha Uzsoy

Published 2026-03-03

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Wolfgang Seiringer, Klaus Altendorfer, Reha Uzsoy

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are the manager of a busy bakery. Every morning, you have to decide when to start baking bread, cakes, and pastries so they are ready exactly when customers walk in.

The Old Way: The Rigid Recipe (MRP)
Traditionally, bakeries use a system called MRP (Material Requirements Planning). It works like a rigid recipe book.

The Rule: "If a customer orders a cake that takes 2 hours to bake, I must start baking it exactly 2 hours before they arrive."
The Problem: This assumes the oven is always free and the baker is always fast. But what if the oven is already full? What if the baker is tired and moving slower?
The Result: If you stick to the rigid 2-hour rule during a busy morning, your cakes will be late. If you try to fix it by starting everything 3 hours early just in case, you end up with a kitchen full of stale, sitting-around cakes (too much inventory). The old system doesn't know how to react to the actual chaos of the kitchen.

The New Way: The Smart Traffic Controller (Clearing Functions)
This paper proposes a smarter way to run the bakery using something called Clearing Functions (CF). Think of this not as a recipe, but as a smart traffic controller for your kitchen.

Instead of saying "Start 2 hours early," the Smart Controller looks at the whole kitchen:

"The oven is currently jammed with 50 loaves of bread. If I start this cake now, it will get stuck in the queue and take 4 hours, not 2."
"However, the mixer is currently empty. I can start the cake batter now, but I'll hold off on putting it in the oven until the bread is done."
"The oven will be free in 30 minutes, so I'll start the cake then."

The Core Innovation: One Size Does Not Fit All
The biggest breakthrough in this paper is that the Smart Controller doesn't give every order the same "lead time" (the time it takes to finish).

Old Way: "All cakes take 2 hours." (Even if the kitchen is empty or packed).
New Way: "This specific cake, ordered for 10 AM, will take 2 hours because the oven is free. That other cake, ordered for 10 AM, will take 4 hours because the oven is backed up, so I'll start it earlier to compensate."

How They Tested It
The researchers built two digital "kitchens" (simulations) to test this:

The Small Bakery (PS1): A simple setup with a few machines.
The Factory (PS2): A massive, complex factory with many machines and hundreds of different products.

They simulated two types of days:

Perfect Forecast: They knew exactly how many customers would come.
Chaotic Forecast: Customers changed their minds at the last minute (e.g., "Actually, I need 50 more cakes!").

The Results: Why the Smart Controller Wins

When things are calm: The old rigid system works okay. It's simple and keeps inventory low.
When things get chaotic (High Demand/Uncertainty): The old system fails. It either makes things too early (wasting money on storage) or too late (angry customers).
The Winner: The Clearing Function approach shines when things are messy. Because it calculates a unique start time for every single order based on how crowded the machines are right now, it:
- Reduces Tardiness: Fewer late orders because it starts busy orders earlier.
- Reduces Waste: It doesn't start everything early just in case; it only starts the specific orders that need a head start.
- Handles the "Last Minute" Panic: If a customer changes an order right before delivery, the Smart Controller can instantly re-calculate and say, "Okay, we need to rush this one," whereas the old system is stuck following its rigid rules.

The Catch
The new system isn't perfect magic. It has some rules:

It assumes that items at the same "level" of the recipe (e.g., all the dough mixers) share the same machines. It doesn't fully solve the problem if a machine is used for both dough and frosting at the same time.
It keeps the basic "recipe" (how much to make) the same as the old system; it just changes when to start making it.

The Bottom Line
This paper shows that in a world where demand is unpredictable and machines get backed up, flexibility is king. By replacing a rigid "start 2 hours early" rule with a dynamic system that calculates the exact start time for every single order based on real-time congestion, factories can save money, reduce waste, and keep customers happy. It's the difference between following a static map and using a GPS that reroutes you around traffic jams in real-time.

1. Problem Statement

The paper addresses a fundamental limitation in traditional Material Requirements Planning (MRP) systems: the treatment of planned lead times (PLTs) as deterministic, exogenous parameters that are independent of system workload.

The Core Issue: In discrete parts manufacturing, cycle times are random variables dependent on resource utilization (workload). However, standard MRP uses backward scheduling with fixed PLTs, assuming infinite capacity or static lead times. This creates a disconnect where release decisions determine the workload, which in turn determines the cycle time, yet the planning model ignores this feedback loop.
The Consequence: Under high utilization or demand uncertainty, fixed PLTs lead to either excessive inventory (if PLTs are too long) or high tardiness and stockouts (if PLTs are too short).
The Goal: To improve MRP performance by replacing the backward scheduling step with an optimization model that computes endogenous, order-specific lead times based on capacity constraints and workload, without dismantling the existing MRP structure.

2. Methodology

The authors propose a hybrid approach that integrates Clearing Functions (CFs) into the MRP framework.

A. Conceptual Framework

Clearing Functions (CFs): A CF is a surrogate model describing the relationship between the workload assigned to a resource in a planning period and its expected output. Unlike linear models, CFs are typically concave, reflecting that output increases at a decreasing rate as congestion rises.
Integration Strategy: The authors do not replace the entire MRP logic. Instead, they retain the standard MRP steps (Netting, Lot-sizing, BOM Explosion) but replace the "Backward Scheduling" step with a CF-based optimization model.
Key Assumptions (Restrictions):
1. Items sharing a production resource must reside at the same BOM level (no cross-level capacity modeling).
2. Production lot integrity from the MRP lot-sizing step is preserved (lots cannot be split).
3. Orders must be released in non-decreasing order of their planned receipt dates (no overtaking).

B. Mathematical Model

The optimization model is a Mixed-Integer Linear Program (MILP) solved for each BOM level and production resource.

Objective: Minimize the sum of Finished Goods Inventory (FGI) holding costs, Work-in-Progress (WIP) holding costs, backlog (tardiness) costs, and external capacity costs.
Decision Variables: Binary variables determining the release period for each specific production order.
Constraints:
- Material Balance: Tracks inventory and WIP flow.
- Lot Integrity: Ensures an order is released exactly once.
- Sequencing: Enforces release order consistency with due dates.
- Clearing Function Constraints: Uses piecewise linear approximations to model the relationship between system load ( $L_t$ $L_{t}$ ) and processed load ( $PL_t$ $P L_{t}$ ). The model tests two CF types:
  1. Ideal/Two-segment CF: Represents a hard capacity limit.
  2. Three-segment CF: Adds a middle segment to model congestion effects (loss of efficiency as utilization approaches capacity).

C. Experimental Design

Simulation Environment: Two production systems were simulated using AnyLogic over 200 periods (with 40 warm-up periods) and 10 replications.
- PS1 (Simple): A 3-level diverging BOM with 3 machines.
- PS2 (Complex): A multi-stage, multi-item flow shop with 32 end items and 4 machines per stage.
Uncertainty Modeling:
- Demand: Modeled using a Martingale Model of Forecast Evolution (MMFE) with updates at $\gamma=10$ and $\gamma=1$ periods before delivery.
- Setup Times: Log-normally distributed (CV=0.2).
Parameters Tested: Utilization levels (80%, 85%, 90%), forecast accuracy (perfect vs. imperfect), lot-sizing policies (Fixed Order Period vs. Fixed Order Quantity), and safety stock levels.

3. Key Contributions

Order-Specific Lead Times: The primary contribution is demonstrating that computing PLTs for individual production orders (rather than for BOM items) significantly outperforms traditional MRP. The CF model dynamically adjusts release dates based on real-time capacity availability.
Modular Enhancement: The paper proves that MRP can be enhanced without a complete system overhaul. By swapping only the backward scheduling step, the system gains the ability to manage congestion proactively.
Handling Uncertainty: The study quantifies how CF-based planning adapts to demand volatility and forecast errors, showing superior robustness compared to static PLT approaches.
Risk Pooling Exploitation: In complex multi-level systems, the CF approach effectively leverages risk-pooling effects at shared resources, smoothing demand fluctuations better than MRP.

4. Key Results

The simulation results highlight distinct performance differences based on system conditions:

Under Perfect Forecasts:
- Low Utilization (80%): CF significantly outperforms MRP by releasing orders earlier to avoid future congestion, reducing tardiness.
- High Utilization (90%): MRP sometimes shows lower total costs because its fixed PLTs prevent premature releases that would inflate inventory. CF's tendency to release early creates "cascading" inventory across BOM levels. However, CF maintains near-perfect service levels (>99%).
Under Imperfect Forecasts (Demand Uncertainty):
- CF Superiority: CF consistently outperforms MRP, especially as demand uncertainty increases.
- Mechanism: CF dynamically adjusts release timing (lower Average Planned Flow Time) to absorb demand spikes. MRP, constrained by fixed PLTs, either releases too late (causing tardiness) or increases PLTs/Safety Stock excessively (causing high inventory costs).
- Cost Savings: In PS2 with high uncertainty, CF reduced total costs by up to 1,500 Cost Units (CU) compared to MRP.
Impact of Lot Sizing:
- CF adapts well to Fixed Order Quantity (FOQ) policies, whereas MRP struggles with FOQ under uncertainty, leading to massive cost spikes due to tardiness. CF's flexibility in release timing mitigates the rigidity of fixed batch sizes.
Congestion Modeling: The three-segment CF (modeling congestion) showed that as production variability increases, costs rise, but CF still manages to balance inventory and tardiness better than static approaches.

5. Significance and Conclusion

Practical Implication: The study validates that a "naive" integration of clearing functions into existing MRP logic can yield significant operational improvements, particularly in environments with high demand uncertainty and tight capacity constraints.
Theoretical Insight: It confirms that workload-dependent, order-specific lead times are superior to static, item-level lead times in stochastic environments.
Limitations & Future Work: The current approach is limited by the assumption that resources are shared only within a single BOM level. Relaxing this to allow simultaneous multi-level optimization is identified as a critical future direction.
Overall Verdict: The paper provides a compelling argument for moving beyond static backward scheduling in MRP. By using clearing functions to dynamically align release dates with capacity, manufacturers can achieve a better trade-off between inventory holding costs and customer service levels.