AM-Articles

January 20, 2025

Agentic AI in Asset Management

What is an AI Agent?

An artificial intelligence (AI) agent is a system that performs tasks on behalf of a user or another system by deciding on its actions using available tools. IBM’s definition of AI agents describes them as systems designed to simulate intelligent human behavior by making decisions autonomously.

Also, an AI agent is a software entity that uses generative AI techniques to autonomously perform tasks or make decisions based on input from its environment. These agents are designed to achieve specific goals or objectives, often by interacting with their surroundings or other agents. Key characteristics of AI agents include autonomy, adaptability, collaboration, and the ability to learn from experience.

Elements of an Agent

Agent have abilities

Agents are systems equipped with several key abilities. They can perceive and interpret their environment using sensors, data inputs, or similar tools to create an understanding of their surroundings. Based on this perception and their predefined goals, agents can process information and make decisions by selecting appropriate actions. These actions can influence their environment, such as providing recommendations, controlling systems, or engaging with users. Additionally, agents often have the capability to learn and enhance their performance over time by leveraging historical data, past experiences, outcomes of previous actions, or user feedback.

Agents have skills

Agents possess a range of skills that enable them to operate effectively. They can role-play by adopting specific behaviors and objectives tied to a particular role, maintaining focus on tasks without getting distracted by irrelevant or time-consuming detours. Agents utilize tools to access additional information, enhancing their capabilities while minimizing distractions through effective tool use for better concentration. As part of their role-play, they can collaborate with other agents, sharing tasks and information to achieve shared objectives. Operating within predefined guardrails, agents stay grounded, avoiding hallucinations and maintaining focus. Additionally, their use of short- and long-term memory supports better collaboration and enhances learning over time.

Type of agents

AI agents span a spectrum from simple rule-based systems, such as scripted chatbots, and advanced adaptive systems like self-driving cars and drones. They can be categorized as reactive, responding to real-time changes; deliberative, planning actions based on goals; or hybrid, combining both approaches.

The foundational types of AI agents include:

Simple Reflex Agents: React to the environment without using memory.
Model-Based Reflex Agents: Use internal models to navigate partially observable environments.
Goal-Based Agents: Make decisions aimed at achieving specific goals.
Utility-Based Agents: Choose actions that maximize utility while achieving goals.
Learning Agents: Continuously adapt and improve by learning from past experiences.

Agents in Maximo Asset Management

Using new Agent AI technologies, IBM Maximo can further achieve greater speed and ease of usage with autonomous automation.

Generative AI reduces reliance on specialized skills not generally available in operational teams and is designed to serve the end user without requiring data science expertise.
Generative AI can interact with watsonx using natural language for insights, actions and automation.
AI agents can work together in networks to perform tasks, combine workflows, and automate operational practices.
AI agents can access the real-time state of equipment using Maximo Monitor, can access historical data in Maximo Manage or stream visual data using Maximo Visual Inspection.
AI agents can use tools based on multi-modal foundation models for language, vision and time-series.
AI agents can use the Maximo APIs to interact with Asset Management workflows and create new business objects, like Work Orders.

Generative AI

Generative AI can Search, Analyze, Summarize and Generate content from large quantities of operational data.

Search for similarities, abnormalities, and relations
Make computations for data aggregations or statistical properties
Make summarizations on search results and format results in tables, graphs or text
Generate reports and apply document styling and formatting using desired natural language

Generative AI can perform, like in Work Order Intelligence, leveraging AI capabilities recommend problem codes for work orders using natural language processing (NLP) and machine learning to analyze historical work orders and identify patterns. Based on this analysis, it recommends a set of problem codes when new work orders are created, enhancing accuracy and data consistency in reporting workflows. By automatically suggesting relevant problem codes, it reduces the time and effort needed by technicians or operators to manually classify issues.

Work Order Intelligence with problem code recommendations.

Conversational AI

Conversational AI provides intuitive, natural language interactions that offer users personalized guidance and support. Conversational AI combines the power of GenAI with an intuitive interface that allows users to achieve their goals more effectively and efficiently by providing AI-Assisted Workflows. Gartner say that “GenAI and LLM are revolutionizing Conversational AI, evolving from traditional chatbots to AI-powered interfaces capable of handling more complex queries and tasks while generating human-like responses”.

Conversational AI can use Retrieval-Augmented Generation (RAG) like SiWarex to unleash hidden intelligence within Maximo to improve work productivity and business outcomes while maintaining, if not improving, trust in generative AI. The AI chain of thought process uses SiWarex to optimize the output of a LLM and references a knowledge base outside of its training data sources to generate a response or take action.

Conversational AI in Maximo supporting an Asset Manager to identify and diagnose an asset at risk and creating a work order with a job plan that mitigates the asset risk.

Agentic AI Networks for Asset Management

Agentic AI can perform assistance in multiple key Asset Maintenance use-cases. Agentic AI refers to artificial intelligence systems that exhibit ‘agency,’ meaning they can act autonomously to achieve specific goals. These systems make decisions, and perform tasks based on their environment and objectives.

Maintenance Operations and Diagnosis. Diagnose a problem, create service requests and work orders, create job plans, and automate the reviews of work orders.
Scheduling, Inventory, and Planning. Schedule inspections and work based on maintenance schedules, technician skills and availability, optimize and rebalance schedules, confirm availability of required parts and tools, and recommend purchase orders for missing parts in plan.
CBM, Reliability, and Predictions. Facilitate tasks related to maintenance strategies and their execution. Monitor conditions, create alerts, compute asset performance scores, and identify failure codes.
Safety and Compliance. Ensure safety and compliance in my procedures. Identify incidents, check for regulations and compliance, and enhancing sustainability through emission tracking.
Reporting. Agents can collect data to compute KPIs

Benefits of Agentic AI Workflows

Agentic AI workflows offer significant advantages by increasing efficiency through automation, enabling tasks to run continuously and at scale while reducing time and resource demands. They are highly scalable, adjusting dynamically to growing workloads without requiring additional human input. With AI agents analyzing data in real-time, decision-making becomes more informed, leading to improved outcomes. These workflows are also adaptable and resilient, maintaining efficiency and effectiveness even in changing or disruptive conditions. Agentic AI depends on clear task statements of Agent outcomes and goals. Concept designs has been exploring the agents supporting Asset Management workflows in Maximo.

Conclusions

Operational organizations strive to continuously improve their maintenance strategies and process maturity leading to increased productivity, efficiency, reliability, and profitability.

Improving the organizational maturity shifts maintenance from being a cost to being an investment.

Development of maintenance strategies.
Industry standards and benchmarking.
Continuous measurement and feedback of KPIs.
Products and services for process implementation.
Process improvement as part of the process.

New generative AI technologies with agent networks provide organizations with new tools to automate process changes while reducing the costs of implementing new maintenance strategies.

Work Order Intelligence in Maximo
Learn more about Work Order Intelligence.

Conversational AI in Maximo
Learn more about Conversational AI.

January 20, 2025

IBM Maximo Named Leader in New Green Quadrant Report

mats AM-Articles

IBM was in October 2024 recognized as a market leader in Enterprise Asset Management (EAM) by the independent research and advisory firm Verdantix, with IBM Maximo Application Suite (IBM Maximo) receiving top marks in the new report “Green Quadrant: Enterprise Asset Management Software 2024.”

The report by Hugo Fuller at Verdantix notes IBM Maximo’s “industry-leading functional capabilities” and “seamless integration with performance-related modules.” The report also acknowledges IBM Maximo’s use of GenAI, noting “IBM’s market-leading EAM suite interweaves maintenance procurement and performance with trailblazing AI.”

“AI-driven roadmap with strong ties to watsonx and IBM Research. With the release of Maximo Application Suite 9.0 in 2024, IBM rolled out generative AI (GenAI) functionality. Notable features are an AI assistant that verifies, corrects or suggests asset failure codes within work orders, and an onboarding assistant trained on Maximo product documentation to expedite the implementation process. IBM plans to develop enhancements for Maximo through IBM Research, with R&D in both transformer models focused on natural language processing and time series models via its open-source Granite models.”

Earlier this year, Verdantix also named IBM Maximo a leader in Asset Performance Management (APM) in the report “Green Quadrant: Asset Performance Management Solutions 2024.” IBM is the only vendor to be named a leader in both the APM and EAM reports.

Verdantix
Read the Verdandix Green Quadrant: Enterprise Asset Management
Software 2024 report.

January 1, 2025

Work Order Intelligence

mats AM-Articles, Design Portfolio

Introduction

The new generative AI capabilities Maximo Application Suite (MAS) 9.0 are the next steps in the Maximo AI roadmap.

The 2024 release of MAS 9.0 is putting the focus on optimizing work order efficiency using generative AI powered by IBM watsonx™. The AI design is delivering on an AI strategy to compel business users to start consuming AI in small steps that deliver business value and progressively leads to increased usage at scale.

We are focusing on the business user and the business value.
We are providing value out-of-the-box
We are providing AI infused into adopted workflows and data
We are providing lifecycle support for models with watsonx

Work Order data quality and process automation AI helps automate the work management workflows and better support for decision making in Manage.

Objective

We will deliver improved efficiency, at the scale of thousands of hours, in everyday work order management by providing more accurate failure code recommendations as compared to current achieved by the organization.

User Research

Research objectives

Multiple qualitative user research projects were conducted as part of the asset management AI strategy and product usage studies, involving sponsor user groups and stakeholder interviews to identify business needs and validate proposed designs.

We sought to understand which personas would benefit most from AI-infused data and how generative AI could enhance their efficiency. Additionally, we explored how users perceive and trust AI recommendations, and we defined how trust in AI systems impacts adoption and usage.

Research targeted both internal and external stakeholders, including Maintenance Managers, Asset Managers, Operation Managers, Reliability Engineers, Technicians, and Business Partners.

Conclusions

“Even small gains in accuracy can have an enormous impact to an organization when it can be applied at scale.”

Organizations today are early in their adoption of Gen AI. Their organization lacks the AI governance and structures needed. Operations are focused on achieving short-term benefits and lack of high quality data is a challenge.

Organizations expect significant value outcomes from generative AI:

10% increase in productivity: Access to recommended problem codes and related job plans will enhance fix rates, leading to cost savings and improved work productivity.
10% increase in compliance: AI recommendations should utilize internal guidelines, regulations, rulings, and data.
20% decrease in asset failures: As reliability engineers leave the workforce, generative AI models are needed to bridge skill gaps by leveraging learnings from historical data.

IBM AI Design

The Work Order Intelligence design is based on ‘Carbon for AI’ in the IBM Carbon design system.

Carbon for AI is an extension of the IBM Carbon design system, designed to give AI instances in IBM products a visually and behaviorally distinct identity. Generative AI introduces a range of new responsibilities when designing experiences, particularly regarding trust, transparency and explainability. Carbon for AI is the common design framework for identifying AI-generated content and delivering explainability in IBM products.

Carbon for AI is mandatory when introducing AI capabilities in IBM products. As IBM Software design leaders we are meeting weekly to share scenarios, review UI and unify usage of Carbon for AI.

AI label and styling

The AI label is a component used as an indicator of AI in the user experience. It is intended to be used in any scenario to identify context generated by AI. It enforces transparency, explainability and feedback. It also becomes the focal point for actions on AI-generated data..

The AI label components.

Explainability

The AI label contains the AI Explainability Popover, which provides a layer of explainability to the user. It provides a consistent summary of relevant model information and offers the option to dig into more details if needed. This may include an explanation of how the AI capability works, the model confidence, the model used, and its version.

The AI Explainability Popover.

Scenario

The end-to-end Work Order Intelligence scenario crosses over multiple maintenance roles and workflows. The main personas are Maintenance Managers, Maintenance Supervisors, Reliability Engineers, Domain Engineers, and Technicians.

First, new work orders are created by technicians from inspections and maintenance tasks. The technician describes the impacted asset, its location, a short and optionally long description of the problem. If known by the technician, the class of failure is provided, and a specific problem code is recorded.

After the work order has been saved, Gen AI makes an inference and predicts the problem codes that best match the work order description. Additional information is also used in the prediction. The predicted problem codes with the highest confidence are captured.

Maintenance Managers and Maintenance Supervisors use the operational dashboard and work queues to drive the work order review workflows at scale. With Work Order Intelligence, the work orders requiring attention may be identified using the ‘WOs with problem code recommendations’ work queue. The number of WOs in the queue is directly visible on the operational dashboard.

When the Maintenance Manager or Maintenance Supervisor reviews the work order, the problem codes recommended by Gen AI are presented in the work order editor using the AI label. The Maintenance Manager views and chooses the best recommendation. Also, before closing a work order, the Maintenance Manager ensures the problem code is updated and matches any additional information or conditions discovered during the maintenance tasks. This ensures that the quality of historical asset data is continuously improved.

A Reliability Engineer may later review problem code recommendations and provide feedback to the model by identifying and tagging good examples of work orders. Retraining the model with such selected work orders improves its accuracy..

Design

Operational dashboard

The operational dashboard is the one-stop destination for the most critical KPIs and access points to maintenance workflows. Work Order Intelligence design identified the need to create a workflow using a work queue that selects all work orders requiring attention. A work queue, ‘WOs with problem code recommendations,’ should be added to the dashboard for that purpose.

Operational dashboard with Work Order Intelligence work queue.

Work Queue selecting all Work Order with problem code recommendation.

Work Order application

The work order application in Maximo Manage is the user experience for most work order related workflows. The application consists of two experiences; a list view page and a details page. The list page displays all, or a filtered set of, work orders. Work Order Intelligence design suggests a predefined view that filters all work orders with recommendations. The work order editor, or details page, is extends the standard editor by infusing AI generated data. New problem code recommendations are highlighted.

WO application and and editor w/ problem code recommendation.

Problem codes and AI recommendations

Problem code recommendations can be inspected with details on the AI model confidence score. The best-fitting code can be selected for the work order. A user can also override any recommendations as decided.

The Gen AI capability generates new recommendations if the work order information is refined or updated. The recommended code may remain unchanged, or new information may generate an updated code with higher confidence.

Problem code recommendations for a work order.

Model management

Work Order Intelligence uses two separate AI models in its implementation.

Problem code predictions are performed by a classification model trained on high-quality work orders spanning all problem codes. The model uses the work order descriptions to classify the problem codes.
To achieve a large training and testing dataset of work orders, a second large language model uses a selected set of work orders to synthetically generate a larger dataset of work order variants. Improved model accuracy is achieved using the generated synthetic dataset.

Selecting work orders for model training is performed in the Work Order application, typically by a Reliability Engineer. The model configuration, models, and model training actions are managed by the system administrator in the AI configuration application.

Add work order to model training set to improve classification model accuracy.

Business Outcome

On June 25, 2024, IBM announced the availability of the 9.0 release of the Maximo Application Suite. Maximo Manage 9.0 introduces AI intelligence to the Maximo work management process with the new problem code classification capability. Maximo Manage 9.0 builds on a long history of AI capabilities in Maximo Monitor and Maximo Visual Inspection.

Work Order Intelligence AI in Maximo was presented at Maximo World 2024.

IBM.com
Maximo Work Order Intelligence: Where AI meets maintenance excellence.

carbondesignsystem.com
Learn more about Carbon for AI.

August 12, 2024

Maximo Conversational AI

mats AM-Articles

Introduction

Maximo Conversational AI is an evolution of previous chat experiences in the Maximo Application Suite. Maximo Conversational AI combines the power of GenAI with IBM Watsonx, IBM Research AI innovations, and the IBM AI conversational assistant interface of IBM Watsonx Orchestrate.

Conversational AI infuses a generative AI prompt experience with situational awareness, providing insights and automation into asset management workflows. Conversational AI also provides new, intuitive, natural language interactions, offering users deep analytic insights into hidden operational and maintenance data.

Using Maximo Conversational AI, new and experienced users can achieve the results of previously used complex queries and scripts by simply using a text prompt. The Conversational AI design concepts expand the use cases from single ad-hoc text prompts to more advanced use cases, allowing the AI system to recognize the context and use previously submitted text prompts as recommended actions or automation agents.

The Maximo conversational AI is based on siwarex, an IBM Research innovation that enables out-of-the-box conversational technologies for the Maximo application management domain. We refer to this concept as Situational Awareness using GenAI. siwarex uses the domain metadata in a knowledge graph for reasoning on prompt input, decomposes and executes the request into parts, and formats the synthesized results.

User research has been performed to identify key business values and achieve trust in AI conversations by business users.

Why conversational AI and automation?

Use GenAI to unleash hidden intelligence within Maximo to improve work productivity, operational resilience and business outcomes.

New modern ways of working: Unleash new ways of working with intuitive, natural language interactions, offering users deep insights into operational data.
Reduce cost: Improve time to value of GenAI, giving time back every day by providing insights and automation into asset management workflows
Identify and apply best practices: Assist in asset diagnosis and quickly identify most relevant prognosis and repair procedures.
Build skills: Reduce knowledge silos & gaps due to an aging workforce by using decades of operational failure, repair and maintenance procedural data.

Objective

Conversational AI capabilities may support a user new to the Maximo suite to enter a text prompt and receive deep insights from decades of operational and maintenance data in a fraction of the time and effort required to write a currently required query or script. We also explore the concepts that allow a user to pin a reusable prompt with context, so that Conversational AI starts suggesting the best-recommended actions in similar conditional situations, reducing any user workload to a single click.

User Research

Research objectives

“I don’t know what to ask the chat to get my job done properly.”

To design a Conversational AI experience for Maximo we needed to uncover customer business needs, their expectations regarding a chat capability, and their willingness to use generative AI to address business problems like reducing effort, cutting costs, and improving operational resilience.
We also need to understand the gap between operational need and skills to use AI chatbots efficiently by technicians in the field.

Conclusions

“My team of operators and technicians need to retrieve the correct information quickly and effectively from the vast sources of operational data, like asset status and health, past incidents, operator logs, maintenance and repairs, to get their work done without costly repair delays.”

Generative and conversational AI give users information faster than ever before, but that convenience is only valuable when it’s built on a foundation of trust to data and action.

Relevance to business value
Coverage of operational data
Data privacy and Cybersecurity

User research found that organizations are cautious about the AI response quality, where a change to a prompt wording may alter the response. Organizations also require data privacy, confidentiality of information used, and on-premises storage of sensitive operational data.

IBM AI Design

The Conversational AI design is based on the new ‘Carbon for AI’ design system and the IBM AI chat conversational framework. Carbon for AI is an extension of the Carbon system, designed to give AI instances in IBM products a visually and behaviorally distinct identity. AI chat is a conversational framework between a user and an AI that can aid in creating tasks, finding insights, tracking documents, and more. Carbon for AI is mandatory when introducing AI capabilities in IBM products. As IBM Software design leaders we are meeting weekly to share scenarios, review UI and unify usage of Carbon for AI.

SiWareX

Situational Awareness using Watsonx (SiWareX) is an AI-powered knowledge discovery system that helps unlock new insights and accelerates data-driven decisions with contextualized industrial data.

The system represents the domain metadata in a knowledge graph. This representation can use existing representations delivered out-of-the-box, or create new custom representations using the discovery APIs of an application. For Maximo, the domain metadata model contains assets, work orders, and users to meet the primary key use cases discovered in user research. The knowledge graph used by SiWareX in preprocessing, is for reasoning on and validating the prompt input. The knowledge graph helps filter an efficient subgraph for the request. By validating consistency on entity names and properties, the system can guide, correct, and prevent the LLM from hallucinating. SiWareX processes the request into a step-by-step (Chain-of-Thought) execution. Each step is automatically executed leveraging the domain metadata. Execution is performed by using the LLM to translate the domain metadata into database queries such as database SQL or application API calls. The results of the steps are synthesized and formatted to the expected output, for example, as a computational number, a dashboard KPI, a result of a work queue, a data table, or a workflow action.

Conversational AI Scenarios

The conversational AI experience design was focuses on the following user scenarios.

Provide assistance in workflows through a guided, interactive, and hands-on experience for new users during trial and onboarding.
Provide an answer to an ad-hoc question, faster than building a query, creating a work queue, implementing a script, or creating a report or report by navigating screens.
Provide data insights to users that today cannot be achieved in the Maximo UI.
Create a high degree of automation using text prompts for actions and instrumentation.Design

Designing a Conversation Experience

“Infuse a generative AI prompts experience, with situational awareness, providing insights and automation, into asset management workflows.”

The Maximo Conversational AI should be thoughtfully designed to deliver a seamless and effective user experience by integrating with existing asset management workflows. It should function as an assistant, providing users with answers to questions, finding and analyzing information, and suggesting workflow-based actions. Conversations should be highly responsive, delivering quick and accurate feedback. Responses should be kept short and summarized to ensure insights are concise, consumable, and actionable. Additionally, the AI should provide navigation links to more detailed data in Maximo and may use the list pages in Maximo to present tabular information. The AI must also be contextual, understanding and adapting to the user’s intent during the conversation. It can leverage memory from previous interactions to suggest recommended actions, assisting users in achieving their goals effectively and intuitively. The system should continuously learn and refine its interactions to improve outcomes, evolving to assist users in achieving their objectives with increasing efficiency and accuracy.

Assistance through a guided, interactive, and hands-on experience for new users during trial and onboarding.

Designing the Interface Experience

The Maximo Conversational AI is based on the IBM AI standards, services, and components:

Carbon for AI design
AI Chat Component
Watsonx Assistant
Watsonx Orchestration

Examples of Conversational AI Designs

A few examples of prompt and action flows in Conversational AI are shown in the image below.

Example of Carbon design language for Conversational AI.

For new users, the conversation assists by suggesting exploring the capabilities of Assets, Work Orders, and Work Planning. Guidance is provided through data templates for bulk creation. For more experienced users, the conversation performs an analysis and identifies assets with high risks, and the common problem code. As an advanced case, a diagnosis is performed and details about an asset are collected and presented.

Maximo Conversational AI Preview Design

The Maximo Conversational AI preview was announced at Maximo World 2024. The implementation is built on Maximo Manage 9.0 baseline. The design for Maximo Conversational AI preview uses floating chat components.

Reduce Potential Failures on Business-Critical Assets

High-priority assets are at risk for failure when they have incomplete and overdue work orders. As a maintenance manager, I need to know which of my high-priority assets have overdue work so that I can start tracking them and remove blockers.

Scenario flow:

Find assets – “Show only asset information about high-priority assets at the Bedford site that have incomplete work orders.
Find assets w/ overdue work – “Make that work orders that are overdue.”
Automate and create new work queue – “Create a work queue named Assets with overdue work.”
Work and prioritize overdue work – “Open work queue.”

Improve Work Efficiency and Reliability on Business-Critical Assets

High-priority assets are at risk due to frequent failures, unclassified failure codes, and long completion time for overdue work orders. As a maintenance manager, I need to understand the completion time and resolution for recently completed work orders to help with my planning and asset reliability.

Scenario flow:

Find work orders – “Show work orders, completed this year, for asset number PM-6072 at the Bedford site.”
Find frequent failures – “Which problem codes occur most frequently among these work orders?”
Understand response time – “Among those work orders with problem code ‘CAVITATN,’ show the average completion time.”
Seek opportunities for WO AI intelligence – “Among those work orders, list their work order number, asset number, and long description.”

Business Outcome

Maximo Conversational AI was presented as a preview at Maximo World 2024.

Next-Gen AI Enabling Situational Awareness in Maximo
Maximo World 2024 session on Conversational AI.

carbondesignsystem.com
Learn more about Carbon for AI.

SiWareX
View research paper on SiWare: Contextual Understanding of Industrial Data for Situational Awareness.