Deployment of an Integrated Management of Change Platform Using Gates Concept and Machine Learning Prediction for Effective Change Evaluation: A Case Study

Abstract Management of change (MOC) is a safety process and an essential part of organizations safety framework. This process is mandated by all safety governing systems. It consists of multiple development and approval stages where it is subject to comment and through assessment. The nature of MOC along with other factors of the requests can be used as features to predict the convergence of the requests to avoid future re-work which results in loss of efforts and manhours if the requests fail to meet the minimum requirements of QA/QC. Considering needs to adapt to economic conditions and market trends, technical development and their evolution change is inevitable in industry, and it is inherent part of the business however to achieve these changes safely and efficiently the MOC process has multiple standardized stages to capture any risks or area for improvement in the potential change. This poses several challenges includes Requests stagnations, missing risk assessments, missing or bypassing steps, underestimating or over estimating costs and schedules. This paper presents a case study of developing a digitized MOC process, covers the design considerations to assure the smooth and efficient workflow of MOC is achieved along with incorporating machine learning predictive capabilities to help predicting scope, cost and schedule compliance level prior authorizing change and focus the review on potential area of improvement. Considering the importance of the development of the deployed MOC platform the process of the design and implementation followed the agile methodology through multiple iterations of setting requirements, deploying the design test and validation and re-adjust the steps and fine tune the requirement on iterative approach. The followed iterative approach enabled the enhancement and streamlining the process to achieve maximum efficiency prior lunching the platform. The main strategies of the devlopment included: Cover the 6 key steps and eliminate repetition and stagnation.Design the process to have each MOC requests pass multiple gates till closure to assure each step are met prior moving forward.Utilize the existing MOC records as training data for a prediction model.Include primary QA/QC steps in the process.Include ML algorithms to support decision making in review and approval cycle. As part of the deployment the key aspects of MOC scope definition identified and engineered which includes Durability, Category, Risk Assessment (methodology), Planned cost, Planned Execution Duration, Shutdown requirement, Number of Unit-to-Unit interfaces Material requirement… etc. The ultimate qualitative outcome of this analysis is predicting whether the proposed change is sanctioned to proceed. The data were pre-processed and included data cleaning to code and correct missing inputs, feature engineering involves creating new variables or features from the existing data to capture complex relationships or enhance the predictive power of the data. For instance, deriving new metrics to quantify employee engagement levels or change readiness can provide valuable insights for change management analysis. Several models were utilized and below is the outcomes of the tested models: According to the outcomes of the modelling the SVM was selected to be the advisory basis of the system to predict MOC convergence and compliance or not for both submitters and reviewers. In conclusion, the deployed design of MOC WF helped in streamlining the process and address the key challenges of the common MOC shortcomings this is attributed mainly to following gate-based WF so no steps can be skipped moreover the applied review and approval constraints eliminated the root causes of requests stations. The incorporation of predicative capabilities helped in avoiding/ minimizing potential requests re-work as it predict the convergence and guide submitter for potential reviewer's concerns in addition to maximize the efficiency of the review cycle as the reviews will be guided to potential failures in term of schedule, scope or cost definitions. The best modeling algorithms for MOC process definition, scope and schedule compliance prediction based on the selected average is found to be support vector machine with accuracy of 93%. The deployed system helped in eliminating the overlapping steps and streamline the process in addition to achieve +98% digitalize MOC control and tracking where predictive capabilities increase MOC convergence by + 30%.