AM Datasets for GenAI Models

Dataset Repository Overview

This section provides a collection of diverse AM datasets that can be used for training and benchmarking GenAI models. These datasets include design, pre-process, build and monitoring, post-process, and quality measurement data, which are essential for developing robust and accurate GenAI models. By leveraging these datasets, researchers can enhance model performance, improve predictive capabilities, and drive innovations in AM.

Each dataset is provided with documentation on its structure, content, and recommended applications. For researchers and practitioners implementing GenAI in AM workflows, these resources provide a solid foundation for model development and evaluation.

Benchmarking Datasets

Benchmarking datasets are specifically designed to evaluate and compare the performance of different GenAI models on standardized AM tasks. These datasets include reference problems, expected outputs, and evaluation metrics to ensure consistent assessment.

Using Benchmarking Datasets

When utilizing these benchmarking datasets, consider the following best practices:

• Use the provided evaluation metrics for consistent comparison
• Report all relevant model parameters and training details
• Benchmark against the provided baseline model performances
• Consider both quantitative metrics and qualitative assessments
• Document any preprocessing or modifications to the dataset

Training Datasets

Training datasets are comprehensive collections designed for fine-tuning GenAI models for AM-specific tasks. These datasets include diverse examples that help models learn the particular characteristics and requirements of additive manufacturing processes.

Datasets for Agents

These specialized datasets are designed for developing and evaluating GenAI agent-based systems in AM environments. They include structured formats for agent interactions, decision-making scenarios, and evaluation frameworks.

Agent Evaluation Frameworks

In addition to the datasets, we provide evaluation frameworks for assessing agent performance in AM contexts:

• Decision Quality Metrics: Evaluating the effectiveness of agent decisions against expert benchmarks
• Interaction Quality Assessment: Measuring the clarity and usefulness of agent communications
• Process Optimization Evaluation: Assessing improvements in manufacturing outcomes
• Adaptation Capability Testing: Measuring how effectively agents adapt to novel scenarios

Prompt Collections

This section provides guidelines and collections for structuring effective prompts when working with GenAI models in AM contexts. Well-designed prompts are crucial for getting desired outputs from GenAI models.

Prompt Engineering Techniques

The prompt collections include examples of five popular prompt engineering methods applied to AM-specific tasks:

• Zero-shot: Prompts that elicit responses without providing examples
• Few-shot: Prompts that include examples of expected outputs
• Chain-of-thought: Prompts that guide the model through a reasoning process
• ReAct: Prompts that combine reasoning and actions in a structured format
• Directional Stimulus: Prompts that guide the model toward specific types of responses

For more information on effective prompt engineering for AM, see our Prompt Engineering Tutorial.

Contributing Data

The AM dataset repository is a community resource that welcomes contributions from researchers and practitioners. By sharing your datasets, you can help advance the field of GenAI in additive manufacturing.

Contribution Guidelines

To contribute datasets to the repository, please follow these guidelines:

• Provide clear documentation of dataset structure and content
• Include information on data collection methodology
• Specify any licensing or usage restrictions
• Remove any sensitive or proprietary information
• Include suggested applications and example usage