MTA-IT-001 - Revision history

Colton Smith: Revision 0 Reviewed by Monica Hurley

2026-03-17T12:38:47Z

Revision 0 Reviewed by Monica Hurley

@@ Line 1: / Line 1: @@
 {{DISPLAYTITLE:Reduce Administrative Overhead through Process Automation using Machine Learning and Artificial Intelligence - MTA-IT-001}}
 {{MTATemplate||
 | Date |12/15/2020
@@ Line 7: / Line 8: @@
 Corrective Action Program
 | Reference Implementation Guidance | DNP-TIP-2020-09, Process Automation Using Machine Learning
-| Industry SME | EPRI - Chris Wiegand
+| Industry SME | EPRI Operating Plant Initiatives
 Contact: NuclearPlantMod@epri.com
 | Previous Implementation | ML and AI has been used at commercial nuclear plants. Contact EPRI for specific implementation examples and utility SME contact information.
 | Implementation Enablers | N/A
 | Applicability | All reactor types
 All geographic regions
@@ Line 63: / Line 57: @@
 * If AI or ML replaces human employees, a plant may not have a human back‑up to the computer system in case of issues. Plants should retain a minimum number of staff trained in critical plant processes.
 * In rare cases, available ML or AI models may not work as expected. These issues may be identified after significant development time has been invested and could lead to development costs without a return. This risk can be mitigated by sharing implementation experience across the industry.

Jacob Sommer: 1 revision imported

2026-03-05T01:37:54Z

1 revision imported

← Older revision	Revision as of 01:37, 5 March 2026
(No difference)

10.142.236.18 at 22:58, 20 February 2026

2026-02-20T22:58:19Z

New page

{{DISPLAYTITLE:Reduce Administrative Overhead through Process Automation using Machine Learning and Artificial Intelligence - MTA-IT-001}}
{{MTATemplate||
| Date |12/15/2020
| Functional Area Where Benefits Will Be Realized | Information Technology
Work Management

Corrective Action Program
| Reference Implementation Guidance | DNP-TIP-2020-09, Process Automation Using Machine Learning
| Industry SME | EPRI - Chris Wiegand
Contact: NuclearPlantMod@epri.com
| Previous Implementation | ML and AI has been used at commercial nuclear plants. Contact EPRI for specific implementation examples and utility SME contact information.
| Implementation Enablers | N/A
| SWEEP Score |
* Cost – Level 3 – The cost per application is less than $1 million.
* Savings – Level 1 – Savings vary by application but are typically less than $1 million per year per application.
* Payback – Level 2 – The payback period for each application is typically between one and five years.
* Licensing Readiness – Level 3 – Impact on licensing depends on the process to be automated. No licensing changes have been required for the tasks in the Description.
* Technology Readiness – Level 3 – Process automation has been implemented at nuclear plants.
* Implementation Proficiency – Level 2 – Process automation requires software interfaces to the plant’s administrative network. Level 2 proficiency is suggested to oversee vendor solutions. A higher degree of proficiency is required to develop in-house solutions.
| Applicability | All reactor types
All geographic regions
| Keywords | Process automation; machine learning; artificial intelligence; AI
| Business Case Analysis Cross-Reference | N/A
}}

==Description==
Many important processes at nuclear plants, such as evaluation of condition reports, tracking and trending, and planning and forecasting, are performed on a routine basis by trained staff, often with years of experience. Staff retirements and workforce reductions increase the burden associated with these processes for the remaining plant staff. Process automation through machine learning (ML) and artificial intelligence (AI) can be used to reduce or eliminate human involvement in these processes.
This MTA discusses such process automation based on an example implementation at a US nuclear plant. The plant developed software and trained the ML‑developed models in‑house to fully or partially automate different processes, including:
* Condition report trend coding
* Duplicate work item detection
* Automatic identification of process parameter anomalies at the plant system level
* Nuclear‑specific part supply and demand forecast
* Identification of common causes
* Automated and preemptive elevated risk assessment of work orders
* Automated screening and processing of condition reports
* Improved scheduling for outage and online work through improved estimation of work duration

The selection of processes was driven by business case analyses and the skill of the project team. A separate application is typically developed to automate each process, although some code may be shared by multiple applications.

==Benefits==
===Benefits Estimate===
Level 1 – Savings in the hundreds of thousands of dollars per year have been reported at a US nuclear plant. Savings greater than $1 million are expected as the program matures and is expanded. Note that savings vary depending on the processes selected for automation.

===Benefits Description===
* Reduced cost and burden for tasks that are automated (or augmented) using ML and AI.
* Improved insight into plant health and processes through data analysis (e.g., process parameter anomalies).
* Improved consistency in the decision‑making process by removing varying personal biases from the process.
* Eliminated dependency on software vendor and reduced costs by in‑house software development and use of open‑source software.

==Costs and Schedule==
===Cost===
Level 3 – The cost per application is typically less than $1 million. For in‑house development, costs are largely the level of effort of the internal development staff. A US nuclear plant invested $500,000 to $800,000 per year over 3.5 years to automate and augment the plant processes listed above. Vendor solutions are estimated to cost approximately $750,000 per application.

===Schedule===
Six months to one year (per application) – Schedule is driven by the complexity of the application and can vary from 2 months to 1.5 years. For in‑house development, time to implementation is dependent on developer expertise and will decrease with experience.

===Scope Context===
Per application

==Risks==
* Computer‑based decisions may be sensitive to small changes in the inputs. To reduce this risk, algorithms can be developed to evaluate these changes and assign a confidence score to the result to reflect uncertainty.
* Computer‑based decisions may not be conservative. To mitigate this risk, the application can be used to pre‑populate form fields for subsequent human review. This approach can be used during the software’s training phase, or permanently. A human review can also be assigned as needed based on the confidence score described above. The risk can also be mitigated by prioritizing automation of form fields and processes that are considered low risk.
* If AI or ML replaces human employees, a plant may not have a human back‑up to the computer system in case of issues. Plants should retain a minimum number of staff trained in critical plant processes.
* In rare cases, available ML or AI models may not work as expected. These issues may be identified after significant development time has been invested and could lead to development costs without a return. This risk can be mitigated by sharing implementation experience across the industry.