Search

LEARNING SURROGATE PRODUCT LIFE CYCLE ASSESSMENT ON ARTIFICIAL NEURAL NETWORK ARCHITECTURE

Benefits of using learning surrogate LCA models


An artificial neural network (ANN) is trained to generalize on characteristics of product concepts typically known in the conceptual design phase, and environmental data from pre-existing LCA. This is done without the overhead of defining new LCA models on a product-by-product basis.

The lack of analytically based methods for incorporating environmental aspects into product concepts motivated the development of the learning surrogate LCA concept. The approach facilitates an integrated system design process, allowing the approximate and rapid assessment of environmental impact based on high-level information typically known in the conceptual phase

The learning surrogate LCA method is appropriate to inform system design decision making in early conceptual stages of product development.

The approach facilitates an integrated system design process, allowing the approximate and rapid assessment of environmental impact based on high-level information typically known in the conceptual phase.

The learning process of the ANN begins when it is provided a set of product descriptors and corresponding detailed LCA results from previously analyzed existing products. The training algorithms adjust parameters within the network so that its output better emulates the actual environmental impact results of the training data products.




Challenges of using learning surrogate LCA models

Life-Cycle Assessment (LCA) is a "cradle-to-grave" approach for assessing the environmental performance of a product, process or service system from raw material acquisition through production, use and disposal.

The aerospace industry currently generates tremendous volumes of data throughout a product life cycle, but the data storage systems are not always designed to have their data extracted, much less at near real-time rates.

The learning surrogate Life Cycle Assessment concept suggests using approximate LCA model based on learning algorithms, learns from existing detailed LCA studies.

The model possesses only a high-level input interface - product concept descriptors - allowing it to operate with the limited data available in early conceptual design stages to meaningfully predict environmental impacts for a wide variety of concepts.

The learning process of the ANN begins when it is provided a set of product descriptors and corresponding detailed LCA results from previously analyzed existing products. The training algorithms adjust parameters within the network so that its output better emulates the actual environmental impact results of the training data products


Implementation of learning surrogate LCA models

The learning surrogate LCA concept is then a powerful alternative to existing approximate LCA approaches. It simultaneously supports:

Life-cycle thinking with lack of detailed information of ill-defined, complex product concept systems.

System outcomes of early conceptual design decisions should be considered. Scarce details on the concepts should no be a barrier to integrate life-cycle environmental assessment early on. The learning surrogate concept overcomes this barrier by bringing into early stages life-cycle knowledge of pre-existing products from which it learns and averages life-cycle performances.

Analysis of substantially different concepts, without the need for building new models. Analytically-based simulation for use in high-dimensional systems and problem specific, multi-attribute trade off-analysis..

Simulation interface between environmental experts and other members of the product design team, in a systems' modelling context.

Environmental issues can be successfully incorporated into early stages of product design only if balanced with the existing traditional design criteria.

Product concept descriptors are a flexible simulation interface between environmental experts and designers for this new approach.

They are also customizable to different high-level parameterized structures that are more helpful in some product design contexts than others, without the need for investing in new explicit modelling tasks



Advantages of using learning surrogate LCA models


Surrogate Life Cycle Assessment method as alternative approach, relies on Artificial Neural Network trained high level product descriptors, without requiring the development of new model, and environmental performance data from pre-existing detailed life cycle assessment studies and related data.

The aerospace industry currently generates tremendous volumes of data throughout a product life cycle, but the data storage systems are not always designed to have their data extracted, much less at near real-time rates. The lack of analytically based methods for incorporating environmental aspects into product concepts motivated the development of the learning surrogate LCA concept

The learning surrogate Life Cycle Assessment concept suggests using approximate LCA model based on learning algorithms, learns from existing detailed LCA studies. The model possesses only a high-level input interface - product concept descriptors - allowing it to operate with the limited data available in early conceptual design stages to meaningfully predict environmental impacts for a wide variety of concepts.



Further remarks using learning surrogate LCA models


The development of the learning surrogate LCA concept facilitates an integrated system design process, allowing the approximate and rapid assessment of environmental impact based on high-level information typically known in the conceptual phase.

An artificial neural network (ANN) is trained to generalize on characteristics of product concepts typically known in the conceptual design phase, and environmental data from pre-existing LCA. This is done without the overhead of defining new LCA models on a product-by-product basis.

The learning process of the ANN begins when it is provided a set of product descriptors and corresponding detailed LCA results from previously analyzed existing products.


The training algorithms adjust parameters within the network so that its output better emulates the actual environmental impact results of the training data products. The process continues until the network converges, or the two outputs - actual and predicted environmental performance - match.

ANNs do not require an explicit functional model for relationships between the system After the completion of training, the ANN is ready for use.


8 views0 comments