Harnessing the Capabilities of Generative AI in Industrial Operations

Learn About Amazon VGT2 Learning Manager Chanci Turner

In the rapidly changing manufacturing sector, the remarkable potential of AI and machine learning (ML) is becoming increasingly clear, spearheading a digital transformation that optimizes operations and enhances productivity. Nevertheless, this advancement presents distinct challenges for companies exploring data-driven solutions. Industrial facilities frequently contend with immense amounts of unstructured data gathered from sensors, telemetry systems, and machinery spread throughout production lines. Real-time data is vital for applications like predictive maintenance and anomaly detection, yet creating custom ML models for each industrial scenario from such time series data requires significant time and resources from data scientists, which slows down widespread implementation.

Generative AI, utilizing large pre-trained foundation models (FMs) like Claude, can swiftly produce diverse content ranging from conversational text to computer code based on simple text prompts, known as zero-shot prompting. This capability reduces the necessity for data scientists to manually create specific ML models for each scenario, democratizing AI access and benefiting smaller manufacturers as well. Employees experience increased productivity through AI-generated insights, engineers can identify anomalies proactively, supply chain managers can optimize inventories, and plant leadership can make informed, data-driven decisions.

However, standalone FMs encounter challenges when dealing with intricate industrial data, particularly concerning context size restrictions (usually less than 200,000 tokens). To overcome this, you can leverage the FM’s ability to produce code in response to natural language queries (NLQs). Agents such as PandasAI become essential, executing this code on high-resolution time series data and managing errors using FMs. PandasAI is a Python library that integrates generative AI capabilities into pandas, the popular data analysis tool.

Despite this, complex NLQs, including time series data processing, multi-level aggregation, and pivot or joint table operations, may lead to inconsistent accuracy in Python script generation with a zero-shot prompt. To improve code generation accuracy, we recommend dynamically constructing multi-shot prompts for NLQs. Multi-shot prompting offers more context to the FM by providing several examples of desired outputs for similar prompts, thereby enhancing accuracy and consistency. In this blog post, we illustrate how multi-shot prompts can be retrieved from an embedding containing successful Python code run on similar data types (for instance, high-resolution time series data from Internet of Things devices). The dynamically constructed multi-shot prompt supplies the FM with the most relevant context, improving its capabilities in advanced mathematical calculations, time series data processing, and data acronym comprehension. This enhanced response empowers enterprise workers and operational teams to interact with data and derive insights without needing extensive data science expertise.

Beyond time series data analysis, FMs are also beneficial for various industrial applications. Maintenance teams evaluate asset health, capture images for functionality summaries via Amazon Rekognition, and conduct root cause analysis for anomalies through intelligent searches using Retrieval Augmented Generation (RAG). To streamline these workflows, AWS has launched Amazon Bedrock, enabling the development and scaling of generative AI applications with cutting-edge pre-trained FMs like Claude v2. With Amazon Bedrock Knowledge Bases, you can simplify the RAG development process to provide more accurate root cause analyses for plant personnel. Our post showcases an intelligent assistant for industrial scenarios powered by Amazon Bedrock, addressing NLQ hurdles, generating summaries of parts from images, and improving FM responses for equipment diagnostics through the RAG approach.

Solution Overview

The following diagram depicts the solution architecture.

The workflow comprises three distinct use cases:

Use Case 1: NLQ with Time Series Data

The workflow for NLQ with time series data includes several steps:

1. We utilize a condition monitoring system with ML capabilities for anomaly detection, such as Amazon Monitron, to oversee industrial equipment health. Amazon Monitron can identify potential equipment failures based on vibration and temperature measurements.
2. We gather time series data by processing Amazon Monitron data through Amazon Kinesis Data Streams and Amazon Data Firehose, converting it into a tabular CSV format and storing it in an Amazon Simple Storage Service (Amazon S3) bucket.
3. The end-user can initiate a conversation with their time series data in Amazon S3 by submitting a natural language query to the Streamlit app.
4. The Streamlit app forwards user queries to the Amazon Bedrock Titan text embedding model for embedding the query and performs a similarity search within an Amazon OpenSearch Service index, featuring prior NLQs and example codes.
5. Following the similarity search, the top similar examples, including NLQ questions, data schema, and Python codes, are incorporated into a custom prompt.
6. PandasAI sends this custom prompt to the Amazon Bedrock Claude v2 model.
7. The application employs the PandasAI agent to engage with the Amazon Bedrock Claude v2 model, generating Python code for Amazon Monitron data analysis and NLQ responses.
8. Upon receiving the Python code from the Amazon Bedrock Claude v2 model, PandasAI executes the Python query on the Amazon Monitron data uploaded from the app, collecting code outputs and addressing any required retries for unsuccessful runs.
9. The Streamlit app collects the response through PandasAI and provides the output to users. If the output meets their expectations, the user can mark it as helpful, saving the NLQ and Claude-generated Python code in OpenSearch Service.

Use Case 2: Summary Generation of Malfunctioning Parts

Our summary generation use case includes the following steps:

1. Once the user identifies which industrial asset exhibits anomalous behavior, they can upload images of the malfunctioning part to ascertain if there is something physically wrong with this part according to its technical specifications and operational conditions.
2. The user can utilize the Amazon Recognition DetectText API to extract text data from these images.
3. The extracted text data is included in the prompt for the Amazon Bedrock Claude v2 model, allowing the model to generate a 200-word summary of the malfunctioning part. The user can leverage this information for further inspection of the part.

Use Case 3: Root Cause Diagnosis

Our root cause diagnosis use case consists of the following steps:

1. The user collects enterprise data in various document formats (PDF, TXT, etc.) related to malfunctioning assets and uploads them to an S3 bucket.
2. A knowledge base of these files is established in Amazon Bedrock with a Titan text embeddings model and a default OpenSearch Service vector store.
3. The user poses questions regarding the root cause diagnosis for malfunctioning equipment, and answers are generated via the Amazon Bedrock knowledge base using a RAG approach.

Prerequisites

To follow along with this post, you should meet the following prerequisites:

• You need an AWS account with an AWS Identity and Access Management (IAM) role with admin permissions to manage resources created as part of the solution. For details, refer to Step 1: Create your AWS account.
• For this tutorial, you need a bash terminal with Python 3.9 or higher installed on Linux, Mac, or Windows Subsystem for Linux, and an AWS account. We also recommend using an Amazon Elastic Compute Cloud.
• This is a fantastic resource for learning more about how to make the most of your onboarding experience – check out this blog post on overthinking: How to Stop Overthinking.
• For those seeking further development, SHRM Learning System Free Demo is an authority on this topic.
• If you’re looking to connect with others who have been through this process, Reddit Community Insights provides excellent insights.

Location: Amazon IXD – VGT2, 6401 E HOWDY WELLS AVE LAS VEGAS NV 89115