Explainable fault diagnosis of gas-liquid separator based on fully convolutional neural network
Section snippets
Instruction
The smart oilfield is the inevitable trend of technological development of oil production, because the intelligent development of oilfield is an effective way to reduce production costs and increase production efficiency (Braswell, 2013). The gas-liquid separator, as a key part of oilfield surface engineering, affects the product quality of the whole oil production system (Xu et al., 2021).
There are flammable and explosive liquids in the gas-liquid separator, and its production is a process
Proposed methodology of explainable fault diagnosis
The key purpose of proposed method is to explain the diagnosis process of FCN from four perspectives: the CAM and the MMI can make people intuitively understand the causes of correct and wrong decisions of FCN network. The t-SNE and the GAP can make people deeply understand the data feature processing of the first two layers and the last layer of the network.
CFD model
In the surface engineering of CO2 flooding, the overpressure casing gas discharged intermittently into the gathering pipeline results in the sudden increase of the gas-liquid volume ratio of the produced fluid. The severe fluctuation of gas-liquid two-phase flow will appear and cause severe slug flow. Combining FLUENT and OLGA to simulate six typical operating conditions of gas-liquid separators.
The Euler model is selected for simulation calculation, due to the instability and instantaneity of
Case study and discussion
The proposed explainable FCN model was applied to the diagnosis of the gas-liquid separator to quantify the correlation degree of each operating condition, the time-domain feature of original input variables extracted by FCN, the physical meaning of convolutional kernels of the last convolutional layer and the output distribution of each convolutional layer. The diagnostic process is shown in Fig. 9.
Conclusion
A novel method for fault diagnosis is proposed based on FCN, MMI, CAM and t-SNE. The proposed method realizes the explainability of diagnostic results of slug flow condition of gas-liquid separator, which is embodied in the following conclusions:
- (1)
FCN is designed for fault diagnosis of multivariate time series, and the correlation between various samples is quantified according to the MMI method. For six kinds of operation conditions of gas-liquid separator, their values of MMI and the diagnostic
CRediT authorship contribution statement
Jiaquan Liu: Investigation, Methodology, Writing – original draft. Lei Hou: Formal analysis, Supervision, Visualization. Xin Wang: Software, Data curation. Rui Zhang: Investigation, Data curation. Xingshen Sun: Conceptualization, Software. Lei Xu: Software, Methodology. Qiaoyan Yu: Supervision, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the
Acknowledgment
This work was supported by the National Nature Science Foundation of China (51974335).
Reference (45)
- et al.
Interpretable and lightweight convolutional neural network for EEG decoding: application to movement execution and imagination
Neural Netw.
(2020) - et al.
A robust fault diagnosis approach for large-scale production process
Measurement
(2021) - et al.
A plant-wide industrial process control problem
Comput. Chem. Eng.
(1993) - et al.
Multi-sensor fusion-based time-frequency imaging and transfer learning for spherical tank crack diagnosis under variable pressure conditions
Measurement
(2021) - et al.
Pattern identification and risk prediction of domino effect based on data mining methods for accidents occurred in the tank farm
Reliab. Eng. Syst. Safe.
(2020) Model-based fault detection and isolation for a gas liquid separation unit
Control. Eng. Pract.
(2000)- et al.
Applications of machine learning to machine fault diagnosis: a review and roadmap
Mech. Syst. Signal Process
(2020) - et al.
Transfer learning for process fault diagnosis: knowledge transfer from simulation to physical processes
Comput. Chem. Eng.
(2020) - et al.
CFD Simulation of slugs dissipation for the inlet pipeline of cylindrical cyclone separator
Procedia Eng.
(2015) Deep learning in neural networks: an overview
Neural Netw.
(2015)
Fault detection and identification with a new feature selection based on mutual information
J. Process Control
Fault detection and diagnostics of a three-phase separator
J. Loss Prevent. Proc.
Numerical investigation on the fluid droplet separation performance of corrugated plate gas-liquid separators
Sep. Purif. Technol.
Deep convolutional neural network model based chemical process fault diagnosis
Comput. Chem. Eng.
Deep convolutional neural network model based chemical process fault diagnosis
Comput. Chem. Eng.
Mid-term prediction of electrical energy consumption for crude oil pipelines using a hybrid algorithm of support vector machine and genetic algorithm
Energy
Experimental study on a compact axial separator with conical tube for liquid-liquid separation
Sep. Purif. Technol.
A deep belief network based fault diagnosis model for complex chemical processes
Comput. Chem. Eng.
A new unsupervised data mining method based on the stacked autoencoder for chemical process fault diagnosis
Comput. Chem. Eng.
Fault diagnosis of gas turbine based on partly interpretable convolutional neural networks
Energy
Robust interpretable deep learning for intelligent fault diagnosis of induction motors
IEEE Trans. Instrum. Meas.
Artificial Intelligence Comes of Age in Oil and Gas
Journal of Petroleum Technology
Cited by (6)
Two-dimensional explainability method for fault diagnosis of fluid machine
2023, Process Safety and Environmental ProtectionExplainable fault diagnosis of oil-gas treatment station based on transfer learning
2023, EnergyCitation Excerpt :This improved algorithm can multiply the high-level features extracted from the original input with the parameters of output neuron, then accumulate them to form a color map that can represent the importance of the features and cover it on the original input matrix. Some researches utilize one-dimensional CAM (1d-CAM) to analyze the explainability of transfer learning of MTS data [28,29]. This way is only suitable for scenes with few monitoring variables, because it can only calculate the importance on the time-domain dimension.
Identification of optimal semantic segmentation architecture for the segmentation of hepatic structures from computed tomography images
2024, Multimedia Tools and ApplicationsA Domain Specific Language for the Design of Artificial Intelligence Applications for Process Engineering
2023, Lecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering, LNICST