asgard-ai-platform Production Planning (MPS/MRP)
Framework
IRON LAW: Plan Hierarchically — Demand → MPS → MRP → Shop Floor
Production planning flows TOP-DOWN:
1. Demand forecast / customer orders → what to make, when
2. MPS → master schedule for finished goods
3. MRP → material and component requirements (what to buy, when)
4. Shop floor scheduling → which machine, which sequence
Skipping levels (going from demand forecast directly to shop floor)
creates chaos — material shortages, capacity conflicts, missed deliveries.
Planning Hierarchy
| Level | Plan | Horizon | Granularity | Decides |
|---|---|---|---|---|
| Strategic | S&OP (Sales & Operations Planning) | 12-18 months | Monthly, product family | Capacity investments, workforce planning |
| Tactical | MPS (Master Production Schedule) | 3-6 months | Weekly, end product | What to produce each week |
| Operational | MRP (Material Requirements Planning) | 4-12 weeks | Daily, component/material | What to order, when, how much |
| Execution | Shop Floor Scheduling | 1-2 weeks | Hourly, work center | Sequence, machine assignment |
MPS Process
- Inputs: Demand forecast + customer orders + current inventory + safety stock targets
- Logic: For each week, calculate: Production Needed = Demand - Beginning Inventory - Scheduled Receipts + Safety Stock
- Output: Planned production quantities by week for each finished product
- Constraint: Must not exceed available capacity (rough-cut capacity planning check)
MRP Process
- Inputs: MPS (what to make) + BOM (Bill of Materials — what it's made of) + Inventory status + Lead times
- Logic: "Explode" the BOM — for each finished good, calculate when each component/material must be ordered to arrive in time
- Output: Planned purchase orders and production orders with dates and quantities
- Key calculation: Order date = Need date - Lead time
Scheduling Algorithms
| Algorithm | Rule | Best When |
|---|---|---|
| FCFS (First Come First Served) | Process in order received | Low complexity, fairness matters |
| SPT (Shortest Processing Time) | Shortest job first | Minimize average flow time |
| EDD (Earliest Due Date) | Most urgent due date first | Minimize late deliveries |
| Bottleneck-first | Schedule the constraint first, subordinate rest | Capacity-constrained environments (TOC logic) |
SMED (Single-Minute Exchange of Die)
Reduce changeover time to increase flexibility:
- Separate internal setup (machine stopped) from external (can do while running)
- Convert internal to external where possible
- Streamline remaining internal steps
- Practice and standardize
Output Format
# Production Plan: {Product/Period}
## Demand vs Capacity
| Week | Demand | Available Capacity | Gap |
|------|--------|-------------------|-----|
| W1 | {units} | {units} | {+/-} |
## Master Production Schedule
| Week | Product A | Product B | Product C | Total |
|------|----------|----------|----------|-------|
| W1 | {units} | {units} | {units} | {units} |
## Material Requirements
| Material | Quantity | Order Date | Supplier | Lead Time |
|----------|---------|-----------|----------|-----------|
| {material} | {qty} | {date} | {supplier} | {days} |
## Scheduling
| Work Center | Mon | Tue | Wed | Thu | Fri |
|------------|-----|-----|-----|-----|-----|
| {center} | {job} | {job} | ... | ... | ... |
Gotchas
- Forecast accuracy limits everything: MPS/MRP is only as good as the demand forecast. Track forecast accuracy (MAPE) and build safety stock proportional to forecast error.
- Lead time variability kills plans: If supplier lead time is "2-6 weeks" (4 week variability), you need safety stock for the worst case. Reduce variability through supplier management.
- BOM accuracy is critical for MRP: A wrong BOM means ordering wrong materials. Audit BOMs quarterly.
- Frozen zone: The first 1-2 weeks of MPS should be "frozen" (no changes) to allow stable execution. Changes inside the frozen zone cascade chaos through the system.
- Finite vs infinite capacity: Basic MRP assumes infinite capacity (ignores machine constraints). Always layer a capacity check (rough-cut or detailed) on top of MRP output.
References
- For S&OP process design, see
references/sop-process.md - For demand forecasting methods, see
references/demand-forecasting.md