Examples of flow optimization by simulation

This page brings together several fictitious case studies designed to illustrate different uses of simulation. The objective is not to describe real projects, but to demonstrate simply and visually how these methods can address various industrial challenges.

01 - Autonomous follower robot (Show)

Less walking, more speed: this case study measures the benefits of a follower robot in order picking. The objective is to quantify the expected operational gains and provide a clear basis for investment decisions.

Description

In a picking area, an operator assembles packages from products distributed across several tables, then seals the package and places it on a conveyor. The main bottleneck observed is the time spent moving, which directly impacts picking speed.

To define the advantages of an Advanced Mobile Robot (AMR), I built a simulation comparing:

the current operation (as is): the operator transports the products and puts them in the package one by one;
target operation (to be): an autonomous mobile robot accompanies the operator and allows him to temporarily deposit the products during collection, in order to limit the back and forth trips.

Results (simulation)

Distance traveled by the operator: 46 m → 26 m (-43.5%)
Preparation time: 52 s/package → 40 s/package (-23.1%)

These gains can inform an investment decision by comparing the operational benefits (reduced travel, increased production rate) with the cost of acquiring the robot.

As is model

To be model

02 - Optimisation (Show)

The goal is to maximise the gain by adjusting the processing times of the two machines connected in series. These common times, shared across all products, are the optimisation levers that enable action on the system's economic performance. In the model below, the optimisation tool integrated into AnyLogic was used to find the best combination of values.

Description

A production line is modelled as a conveyor with four product types (A, B, C, D) that arrive at regular intervals. Each product passes successively through two machines in series.
The goal is to maximise the overall gain by adjusting the processing times of the two machines (decision variables), while also integrating:

an energy cost related to the machines,
a penalty related to the size of the queue,
and a profit per product.

To identify the best settings, AnyLogic's built-in optimisation tool (of the genetic algorithm type) explores different combinations of durations and retains those that maximise the gain.

Results (optimisation)

Current setting: processing time machine 1 = 5 s, processing time machine 2 = 5 s → Gain: 5
Optimized setting: machine 1 processing time = 4 s, machine 2 processing time = 5 s → Gain: 121.6

Recommendation: Adjust the processing time of machine 1 to 4 seconds; the processing time of machine 2 can remain unchanged.

As is model

To be model

03 - Autonomous mobile robot (Show)

This case study shows how a dedicated autonomous mobile robot (AMR) for refuelling can relieve operators of trips to the store, reduce unnecessary travel, and increase production, based on a “as is /to be” simulation comparison.

Description

The workshop has 9 workstations, each occupied by 1 operator. To produce, each operator uses parts from boxes stored in the warehouse.

Current model (as is): the operator must go and get a box from the store himself, then return to his station.
Future model (to be): An AMR is responsible for transporting boxes to a dedicated drop-off area in front of each workstation. It anticipates demand and keeps a box on hand in advance, thereby limiting stockouts and back-and-forth trips.

Results (simulation)

Operator travel time: 450 min → 125 min (-72.2%)
Waiting time / non-production: 450 min → 135 min (-70%)
Average distance covered: 14.1 km → 5.4 km (-62%)
Percentage of time spent commuting: 6–21% → 3% (–50% to –85.7%)
Total boxes processed: 112 → 126 (+12.5%)

Conclusion: the integration of a refuelling RMA significantly improves working conditions (less walking, less waiting) and increases productivity, factors that must be weighed against the cost of setting up the robot.

As is model

To be model

4 - Queue optimisation by artificial intelligence (Show)

This case study illustrates how an artificial intelligence approach can help reduce waiting times in a multi-queue system by making more nuanced customer-orientation decisions than a simple "go in the shortest queue" rule.

Description

The system under study represents a service organisation with 6 queues and 6 service counters (one server per queue). Customers arrive continuously, and their requests can be of several types, with varying processing times.

Two strategies are compared under identical circumstances:

As is model (simple rule): each customer is directed to the shortest queue upon arrival.
To be model (with AI): an AI agent (reinforcement learning, PPO) learns to distribute the load, taking into account the state of the queues and service times, in order to minimise the average waiting time.

Results (simulation)

At the end of the simulation:

Average waiting time — “shortest queue” rule: 8.8 s
Average waiting time — AI strategy: 5.9 s
Improvement: 33%

Conclusion: an AI-driven guidance strategy enables significantly reducing the average waiting time without increasing resources (same counters) by improving load distribution.

As is model (shortest line)

To be model (with AI)