AI warehouses
A new generation of Al-empowered warehouses.
COMPANY
Johnson&Johnson
RESEARCH CENTRE
MIT - Center for Transportation & Logistics
PRODUCTIVE SECTOR
Logistics and Transport
Problem description
In the context of a fully automated warehouse, can Al help us reduce material movements, decrease energy consumption and lower wear and tear on the equipment?
Challenges and goals
The goal of this collaboration is to develop an approach to locate the storage of materials within the warehouse by minimizing the distance the crane travels. The full optimization problem of material location assignment itself is complicated. We first need to predict the material movements and make assumptions to keep the optimization problem more manageable.
Mathematical and computational methods
-
Deep Neural Networks are a type of artificial neural network that consists of multiple layers of interconnected neurons or units where the weights associated with each connection are adjusted to minimize the error between its predictions and the actual target values. Two famous architectures are Feedforward Neural Networks (FNN) and Long-Short Term Memory Networks (LSTMs).
-
Mathematical Optimization is a field of mathematics and computer science that deals with finding the best solution from a set of possible solutions to a particular problem. It requires an objective function, constraints and an optimization strategy.
min f(x)
st. gj (x) ≤ 0, j = 1, ... , p,
hi (x) = 0, i = 1, ...,m,
where gj hi: Rn -> R are inequality and equality constraint functions, respectively. -
Bayesian Optimization is a systematic approach to hyperparameter tuning that uses probabilistic modeling and acquisition functions to efficiently search for the best hyperparameter configuration while minimizing the number of objective function evaluations.
Results and Benefits
The underlying mathematical novelty was to design a two-stage approach that redefines the material categories and their placement within the warehouse. To achieve that, we first leveraged the history of transfers to forecast the amount of monthly movements for each material using two models borrowed from the Deep Neural Networks space: FNNs and LSTMs.
Then we optimized the storage of materials within the warehouse by minimizing the distance the crane travels subject to space, allocation and assignment constraints.
