AI-Driven Predictive Maintenance Boosts Efficiency

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jul 8 2024

In a paper published in the journal Machines, researchers developed a predictive maintenance (PdM) framework to enhance production efficiency and reduce costs. Utilizing Industry 4.0 technologies and machine learning (ML), they focused on performing maintenance only when necessary.

The framework linked machine state variables with product quality (PQ) parameters and was tested on electromechanical component production lines. This approach improved machine reliability, saving about 50% in machine downtime (CostDT) costs and 64% in scrap costs (CostS). The study highlighted the significant impact of PQ on manufacturing costs and reliability.

Background

Past work aimed to guide PdM implementation using ML algorithms based on Industry 4.0 technologies and PQ data. Traditional preventive maintenance (PvM) strategies are costly and often lead to unnecessary actions. In contrast, PdM optimizes maintenance using real-time data to predict failures and plan just-in-time maintenance, improving efficiency and reducing costs. Integrating the Internet of Things (IoT) and ML has advanced PdM strategies, but few studies combine these with PQ data.

Research Methodology

This study investigates the connection between PdM and PQ. The first research question examines how PQ parameters can influence PdM. In contrast, the second explores the potential impact of PdM on improving mean time between maintenance (MTBM) and reducing costs through PQ optimization. The research framework consists of three main steps: defining input variables, developing a predictive analysis model, and validating the results.

Step 1 identifies machine status and PQ variables as input and output parameters. Step 2 involves creating a model that combines machine status monitoring with PQ prediction using various ML techniques, comparing their accuracy to select the best algorithm. Step 3 validates the model's application in real-world scenarios, assessing its impact on maintenance actions, costs, and quality rates.

Predictive Maintenance Implementation

This section demonstrates the application of the research framework to a case study involving a company that produces electromechanical breakers with 10 production lines across five plants. The study focuses on a power transformer assembly line, particularly the vacuum mixer machine, which critically impacts product quality. The process involves assembling cores, soldering, applying resin, drying, and testing.

With its vacuum pumps, the vacuum mixer machine is crucial for ensuring insulation quality, as any deviation in vacuum pressure can lead to defects. PdM was applied to monitor the vacuum pumps, focusing on pressure and temperature parameters to predict and prevent failures, using a combination of ML techniques and a fuzzy inference engine (FIE).

In step 2, various ML techniques (naive Bayes, nearest neighbor, and bagged tree classifiers) were evaluated for predicting first pass yield (FPY) based on machine status inputs, with the bagged tree classifier (BTC) showing the highest accuracy at 91%. The BTC, combined with FIE, helped decide maintenance actions based on FPY predictions and machine parameters. Step 3 validated the model through real-case scenarios, detecting anomalies and suggesting corrective actions. Over six months, the model demonstrated significant improvements in MTBM from 37.8 to 48.6 hours and FPY, enhancing machine availability from 93.1% to 96.6%, thereby confirming the positive impact of integrating product quality parameters into the PdM framework.

Cost Savings Analysis

This section evaluates the cost reduction achieved by applying a PdM model that considers PQ. The total cost before implementing the framework is compared with the result after its application. The total cost (CostTOT) includes CostDT, maintenance cost (CostM), CostS, and the implementation of the artificial intelligence model (CostAI) over a period (usually one year).

The CostDT is calculated based on the missed production hourly cost (K1), the number of maintenance events (NME), and the mean time to repair (MTTR). CostM includes the cost of maintenance events, spare parts, and labor. CostS is determined by the labor cost, production volume, FPY, and reworking time (RWT). CostAI covers the model's depreciation and management costs.

Applying the cost evaluation to the case study over 150 days showed significant cost savings. The model reduced costs related to CostDT by approximately 50% and CostS by 64%. This demonstrates the financial benefits of integrating a PdM framework that incorporates product quality parameters, enhancing both production efficiency and cost-effectiveness.

Conclusion

To sum up, PdM was crucial in optimizing production efficiency by minimizing CostDT and reducing CostS. Unlike traditional preventive methods, it extended component life by continuously monitoring system behavior, reducing failures and unnecessary maintenance. The adoption of Industry 4.0 technologies has significantly enriched predictive maintenance by providing extensive datasets, which empower AI-driven solutions utilizing various ML algorithms. Integrating PQ metrics into this framework enhanced production continuity and quality rates while reducing operational costs.

The framework's novelty lay in its integrated approach to managing process performance, balancing downtime duration, stoppages, and waste. Linking the machine state with production quality using ML techniques facilitated proactive maintenance decisions that optimized operations and cost efficiencies. Applied to a real case study, the framework demonstrated significant cost savings, validating its effectiveness in pre-empting machine failures and optimizing production quality.