Research Paper Example: Development Of A Health Performance Monitoring Tool For Centrifugal Gas Compressor Using K-NN Model

Development Of A Health Performance Monitoring Tool For Centrifugal Gas Compressor Using K-NN Model

1. Abstract

1.1 Summary of objectives and findings

This research presents the development of a health performance monitoring tool for two-stage centrifugal gas compressors using a k-Nearest Neighbor (K-NN) classification model. By analyzing operational data—such as shaft speed, pressures, temperature, and flow—collected over a year, the proposed system achieved an accuracy of 84.4 % in distinguishing healthy from faulty compressor states. These outcomes demonstrate the efficacy of K-NN for predictive maintenance, enabling proactive decision-making to enhance reliability and reduce downtime (Shar et al.).

2. Introduction

2.1 Background and significance

Centrifugal gas compressors are critical in oil and gas operations, offering high flow rates and variable pressure ratios superior to positive displacement machines (Zhang et al. 174). Traditional maintenance strategies often rely on reactive or scheduled routines, leading to costly unplanned shutdowns and inefficient resource utilization (Nsor). The advent of Industry 4.0 has propelled the adoption of data-driven predictive maintenance techniques that leverage artificial intelligence for real-time equipment health assessment (Tsallis et al. 4898).

2.2 Study objectives

This study aims to design and implement a health performance monitoring tool based on the K-NN model to (1) detect anomalies indicating fouling, surging, or bearing faults and (2) provide early warnings for maintenance planning to minimize downtime and optimize energy consumption (Shar et al.).

3. Methodology

3.1 Data acquisition and preprocessing

One year of operational data from a two-stage centrifugal gas compressor—including suction and discharge pressures, temperatures, volumetric flow, and rotational speed—was collected at one-hour intervals. Data were organized by sensor tags and cleaned to remove missing values and outliers, resulting in 5,810 instances for training and 2,119 for testing (Shar et al.).

3.2 K-NN model development

Using Orange software version 3.340, the K-NN classifier was trained on 70 % of the dataset. A trial-and-error approach determined the optimal k value balancing bias and variance. The model labels each instance as healthy or faulty by majority vote among its k nearest neighbors in feature space (Shar et al.).

3.3 Performance evaluation metrics

Model performance was primarily assessed using classification accuracy, aligned with best practices in predictive maintenance studies (Susto et al.). Additionally, an energy performance indicator (EnPI) was derived to detect deviations in energy efficiency, serving as an early anomaly detection metric (Shar et al.).

4. Results

4.1 Model performance summary

The K-NN model yielded an accuracy of 84.4 % on the test set, effectively distinguishing between healthy and faulty compressor states. This performance confirms the model’s capacity to capture nonlinear relationships in sensor data and detect energy performance deviations (Shar et al.).

4.2 Performance comparison graph

Figure 1: Illustrative performance comparison of k-NN and other classifiers. (Data not derived from provided sources)

Note: This section includes information based on general knowledge, as specific supporting data was not available.

Discussion

5.1 Interpretation of results

The 84.4 % accuracy demonstrates that K-NN effectively identifies complex anomalies in compressor sensor data, outperforming traditional threshold-based methods that often miss nonlinear interactions (Shar et al.). The inclusion of EnPI further enhances early anomaly detection by monitoring energy efficiency trends.

5.2 Implications for health monitoring

By enabling proactive maintenance scheduling, the proposed tool reduces unplanned downtime, optimizes energy consumption, and lowers operational costs. Operators gain actionable insights into compressor health, supporting timely interventions and improving overall system reliability (Tsallis et al. 4898).

6. Conclusion

6.1 Key findings and future work

This study validated the use of a K-NN based monitoring tool to achieve 84.4 % classification accuracy in predicting centrifugal compressor health. The approach facilitates a transition from reactive to data-driven maintenance. Future research should explore hybrid models integrating deep learning and IoT-enabled sensors to enhance detection accuracy and scalability (Shar et al.).

7. References

Shar, MukhtiarAli, MehtabAli, KaisarAli, and HamzaAbdulJalil. Development Of A Health Performance Monitoring Tool For Centrifugal Gas Compressor Using K-NN Model. 20XX.

Zhang, H., et al. “Experimental Investigation of Characteristics of Instability Evolution in a Centrifugal Compressor.” Chin. J. Aeronaut., vol. 36, no. 4, 2023, pp. 174–189.

Tsallis, C., P. Papageorgas, D. Piromalis, and R. A. Munteanu. “Application-Wise Review of Machine Learning-Based Predictive Maintenance: Trends, Challenges, and Future Directions.” Appl. Sci., vol. 15, no. 9, 2025, p. 4898.

Nsor, M. “Predictive Maintenance Using Machine Learning for Engineering Systems Through Real-Time Sensor Data and Anomaly Detection Models.” Int. J. Res. Publ. Rev., vol. 6, no. 7, 2025, pp. 5167–5183.

Susto, G. A., A. Schirru, S. Pampuri, S. McLoone, and A. Beghi. “Machine Learning for Predictive Maintenance: A Multiple Classifier Approach.” IEEE Trans. Ind. Inform., vol. 11, no. 3, 2015, pp. 812–820.