Geri Dön

Cloud-based diagnosis of refrigerant leakage fault of chiller using virtual sensor residuals assisted xgboost algorithm

Soğutma sistemlerinin soğutucu akışkan sızıntısı arızasının sanal sensör farkı destekli xgboost algoritması ile bulut tabanlı tespiti

  1. Tez No: 770571
  2. Yazar: BURKAY ANDUV
  3. Danışmanlar: DOÇ. DR. ZHIMIN DU
  4. Tez Türü: Yüksek Lisans
  5. Konular: Fizik ve Fizik Mühendisliği, Makine Mühendisliği, Physics and Physics Engineering, Mechanical Engineering
  6. Anahtar Kelimeler: Deep learning, XGBoost, Preprocessing, HVAC system, FDD, Regression
  7. Yıl: 2022
  8. Dil: İngilizce
  9. Üniversite: Shanghai Jlao Tong University
  10. Enstitü: Yurtdışı Enstitü
  11. Ana Bilim Dalı: Makine Mühendisliği Ana Bilim Dalı
  12. Bilim Dalı: Belirtilmemiş.
  13. Sayfa Sayısı: 127

Özet

Over the recent decades, improvements in the sensor technology have enabled researchers to access extensive operational data of HVAC chillers. This large amounts of data have facilitated the adaptation of novelsmart fault detection and diagnosis (FDD) methods as a means to increasing the energy efficiency of buildings. However, although the data processing methods are very powerful, the experimental data collected from chillers generally lack the sufficient diversity and standardization that the data driven methods require to obtain a general model that can be applied in various cases. A deficiency in preprocessing or a misconception in the selection evaluation metric may easily result in a misinterpretation of the test results. If there is such misjudgment, the final FDD method would suffer heavily from low generalization capacity and overfitting. This paper highlights the origins of common misconceptions and attempts to achieve a higher generalization performance than the conventional methods by constructing a novel procedure to diagnose the refrigerant leakage fault of chillers. To achieve this target, first the faulty and fault-free operation data of an industrial chiller is collected. Then, after conducting the relevant preprocessing steps, an algorithm that creates virtual sensors based on fault-free conditions is constructed. This algorithm is then used to calculate selected residuals of actual sensor readings and virtual sensor predictions. An optimized XGBoost algorithm is supplied with both normal and faulty conditions as inputs to make the final prediction on the refrigerant charge level. The diagnosis results from the proposed method are compared with other widely-accepted methods such as Support Vector Machines (SVM), Multi-Layer Perceptron (MLP), Random Forest (RF) and pure Extreme Gradient Boosting (XGBoost). The results reveal that, all of the traditional methods have above 99% accuracy in random testing. Yet, when these methods are kfold tested, even the best compared model could only achieve accuracies of 54%, 68% and 72%. The proposed method, on the other hand, achieved an exceptional generalization performance with accuracy results of 68%, 70% and 72%, while the method still retained above 99% accuracy in random testing. As the final step, a cloud based FDD framework is constructed to track the status chiller health from a user friendly website. The framework uses a dedicated server for its FDD module, where the software makes judgements on chiller health by predicting the refrigerant load level from the sensor data obtained by the chiller in real-time. The proposed model is uploaded to the framework and the study is concluded with conducting simulations and tests on the online system.

Özet (Çeviri)

Özet çevirisi mevcut değil.

Benzer Tezler

  1. Bulut tabanlı çoklu akıllı kuvöz izleme sistemi tasarımı

    Cloud based multi incubator monitoring system design

    UĞUR ALTAN

    Yüksek Lisans

    Türkçe

    Türkçe

    2019

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİskenderun Teknik Üniversitesi

    Elektrik-Elektronik Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ AHMET GÖKÇEN

  2. Bulut tabanlı kablosuz EEG kayıt sisteminin tasarımı ve gerçeklemesi

    Design and implementation of a cloud-based wireless EEG recording system

    HASAN BİTAR

    Yüksek Lisans

    Türkçe

    Türkçe

    2022

    Elektrik ve Elektronik MühendisliğiOndokuz Mayıs Üniversitesi

    Elektrik-Elektronik Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ CENGİZ TEPE

  3. Real-time performance diagnosis and evaluation of big data systems in cloud datacenters

    Bulut veri merkezlerinde büyük veri sistemlerinin gerçek zamanlı performans teşhisi ve değerlendirilmesi

    ÜMİT DEMİRBAGA

    Doktora

    İngilizce

    İngilizce

    2021

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolNewcastle University

    Bilgisayar Bilimleri Ana Bilim Dalı

    PROF. DR. RAJIV RANJAN

  4. CNN-based server state monitoring and fault diagnosis using infrared thermography

    Kızılötesi termografi kullanarak CNN tabanlı sunucu durumu izleme ve arıza teşhisi

    BELTUS NKWAWİR WİYSOBUNRİ

    Yüksek Lisans

    İngilizce

    İngilizce

    2022

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Teknik Üniversitesi

    Bilgisayar Bilimleri Bilim Dalı

    DR. ÖĞR. ÜYESİ HAMZA SALİH ERDEN

    PROF. DR. BEHÇET UĞUR TÖREYİN

  5. İdiopatik intrakranial hipertansiyon hastalarında serebral anatomik yapıların mrg volumetri ile değerlendirilmesi

    Assessment of cerebral anatomic structures using mr volumetry in patients with idiopathic intracranial hypertension

    REVZA YALÇIN

    Tıpta Uzmanlık

    Türkçe

    Türkçe

    2023

    NörolojiAtatürk Üniversitesi

    Nöroloji Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ RECEP YEVGİ