Surge prediction method and system for centrifugal compressors

Abstract

The present disclosure provides a computer-implemented method for predict- ing surge in equipment, such as a centrifugal compressor. The method comprises re-5 ceiving a plurality of operating data from the equipment, predicting a surge distance using at least one machine learning model, where the surge distance is calculated as the difference between the current flow rate and the predicted flow rate at surge, com- paring the predicted surge distance to a predetermined threshold, and generating an alert signal when the predicted surge distance is below the predetermined threshold. 10 This method anticipates and mitigates potential surge events, ensuring optimal equip- ment performance and safety. The present disclosure also discloses also a system for predicting surge in an equipment. 15

Description

Surge Prediction Method and System for Centrifugal CompressorsDescriptionTECHNICAL FIELD

[0001] The present disclosure concerns a method for predicting the surge of a centrifugal compressor and the relevant system for implementing and executing the same.

[0002] More specifically, the present disclosure relates to surge prediction systems for compressors, particularly a real-time surge prediction system for centrifugal compressors, using machine learning models trained on simulated data.BACKGROUND ART

[0003] Centrifugal compressors are widely used in various industrial applications, including oil and gas, chemical processing, and power generation. These compressors increase the pressure of gases for processes such as gas transmission, refrigeration cycles, and petrochemical production.

[0004] However, centrifugal compressors are susceptible to a phenomenon known as surge, which can occur when the compressor operates below a certain flow rate for a given rotational speed.

[0005] Surge is an unstable operating condition characterized by rapid flow reversals and pressure fluctuations within the compressor. This instability can lead to severe mechanical stress, reduced efficiency, and potential damage to the compressor and associated equipment.

[0006] Preventing surge events is particularly important, for maintaining safe and efficient operation of centrifugal compressors.

[0007] Currently, surge prevention has relied on maintaining a safe operating margin between the compressor’s operating point and the surge line. This approach often results in conservative operation, necessarily reducing efficiency and the overall performance of the machine.

[0008] More advanced methods have attempted to detect the onset of surge through various means, such as monitoring pressure fluctuations or vibrations. However, these methods typically react to imminent surge conditions rather than predicting them in advance, thus actually not solving the problem.

[0009] Recent advancements in data analytics and machine learning have opened new possibilities for more accurate and proactive surge prediction. These techniques have the potential to leverage large amounts of operational data to identify patterns and trends that may indicate an increased risk of surge.

[0010] However, implementing such systems in real industrial environments presents hurdles, including the need for extensive historical data, details of the complexity of compressor behavior under several and varying operating conditions, and the requirement for real-time processing capabilities, which are available only with recent technologies.

[0011] The prior art comprises the US Patent Application No. 2017 / 0370368 Al, which discloses a system and method for predicting surge events in compressors of turbomachines. This patent application describes the use of model parameters and historical data to analyze compressor efficiency and predict surge events.

[0012] While the above approach represents a step forward in surge prediction, there remains a need for more accurate and responsive systems that can operate effectively across a wide range of compressor designs and operating conditions.

[0013] As industrial processes continue to demand higher efficiency and reliability from centrifugal compressors, the development of more sophisticated surge prediction systems becomes increasingly important. Such systems must be capable of providing accurate, real-time predictions while adapting to the characteristics of specific compressors.

[0014] The relevant prior art also comprises the patent applications US2020 / 284265A1, US5873257A and CN109611370A.SUMMARY

[0015] In one aspect, the subject matter disclosed herein concerns a computer-implemented method for predicting surge in an equipment. The method comprises receiving a plurality of operating data of the equipment by processing means, predicting a surge distance using at least one machine learning model, comparing the predicted surge distance to a predetermined threshold, and generating an alert signal when the predicted surge distance is below the threshold.

[0016] In another aspect, disclosed herein is a method wherein the receiving step includes adjusting the received operating data to obtain adjusted data. The operating data may include parameters such as inlet pressure, inlet temperature, outlet pressure, outlet temperature, mass / volumetric flow rate, gas composition, and rotational speed.

[0017] A further aspect of the present disclosure is drawn to a method that further includes predicting a mass flow rate using the machine learning model and verifying the accuracy of the model by comparing predicted and measured outlet pressures and temperatures.

[0018] In another aspect, disclosed herein is a method where the machine learning model is trained using a procedure based on simulated data, which involves generating the simulated data via simulation software and calculating model parameters during the training. The method also includes storing performance parameters and predicted surge distances in a database, and retraining the model using stored data to enhance prediction accuracy over time.

[0019] A further aspect of the present disclosure concerns a method in which the machine learning model could be a Polynomial Ridge Regression, Gaussian Process Regression, Deep Neural Network, or Gradient Boosted Tree Regression, among others.

[0020] In another aspect, the disclosed method comprises the use of sensors for acquiring operating data, and processing means connected to these sensors. The processing means may comprise a first processing unit for executing the training procedure and a second processing unit for executing other method steps. The equipment could be a centrifugal compressor.

[0021] In another aspect, disclosed herein is a system for predicting surge in an equipment, comprising sensors to measure operating data, processing means connected to the sensors, and a storage unit, where the system is configured to execute the aforementioned method.BRIEF DESCRIPTION OF THE DRAWINGS

[0022] A more complete appreciation of the disclosed embodiments of the invention and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:Fig. 1 illustrates a block diagram of a surge prediction system for a centrifugal compressor, according to aspects of the present disclosure;Fig. 2 illustrates a flowchart of a method for predicting surge in an equipment, according to an embodiment;Fig. 3 illustrates a flowchart for additional steps in a surge prediction process, according to aspects of the present disclosure;Fig. 4 illustrates a flowchart for additional steps in a surge prediction process, according to aspects of the present disclosure;Fig. 5 illustrates a flowchart for additional steps in a surge prediction process, according to aspects of the present disclosure; andFig. 6 illustrates a graph showing the relationship between QI and P2 / P1 for a centrifugal compressor, according to an embodiment of the present disclosure.DETAILED DESCRIPTION OF EMBODIMENTS

[0023] According to one aspect, the present subject matter is directed to a system and method for predicting misfunctions in equipment, and particularly surge events in centrifugal compressors. The solution disclosed utilizes machine learning models trained on simulated data to predict a measure of the likelihood of a surge, called surge distance, in real-time, offering a proactive approach to surge prevention. The system is designed to adapt to varying operating conditions, taking into account factors such as different gas compositions and machine component sizing. This adaptability allowsthe system to be applicable across a wide range of industrial applications and compressor (or equipment in general) designs.

[0024] The system also comprises machine learning models for predicting various model parameters, such as mass flow rate at choke, outlet pressure, and temperature, providing a comprehensive assessment of compressor performance.

[0025] Furthermore, the system is capable of continuously improving its prediction accuracy through retraining with newly acquired operational data, ensuring that the system becomes increasingly accurate to the specific characteristics of the compressor or equipment it monitors overtime. The real-time processing capabilities of the system allow for immediate response to changing operational conditions, a crucial factor in preventing surges in industrial environments.

[0026] Referring to Fig. 1, a surge prediction system 1 for a centrifugal compressor 2 is illustrated.

[0027] The operation of the surge prediction system 1 can be divided in two main environments or moments: an offline environment and an operating environment.

[0028] In general, the surge prediction system 1 comprises a first processing unit 31, and a storing unit 32, connected or associated to the first processing unit 31, and a second processing unit 33, connected to, or installed in the centrifugal compressor 2, whose operation will be better defined below. In one embodiment of the invention the first processing unit 31 and the second processing unit 33 can be the same processor unit.

[0029] In the offline environment, the first processing unit 31 is programmed to carry out simulations on the centrifugal compressor 2 to train one or more machine learning models 4 of the equipment to be monitored, namely, in the present embodiment, the centrifugal compressor 2.

[0030] In the foregoing we will refer to a machine learning model 4, but it is intended that also multiple machine learning models can be considered or deployed.

[0031] The first processing unit 31 of the surge prediction system 1 uses simulateddata to train the machine learning model 4, wherein such simulated data may be generated using a simulation software based on analytical empirical formulas, maps, and numerical models specific to the centrifugal compressor 2 or the equipment in general.

[0032] The simulation software may be capable of simulating the impact of different operating conditions (gas compositions, pressure and temperature) and machine component sizing on the performance of the centrifugal compressor 2. This allows the machine learning model 4 to be trained on a wide range of operating conditions, improving its ability to predict surge the operation and potential harmful events under various scenarios.

[0033] By the simulations carried out, the first processing unit 31 generates a set of model parameters specific of the centrifugal compressor 2 to be monitored, to make the machine learning model 4 operative and predictive of the operation of the specific equipment. Such model parameters may be stored in the storing unit 32.

[0034] In some embodiments, the first processing unit 31 may be implemented as a single unit or as separated parts, namely a local server, for managing and storing the model parameters deriving from the simulated data, and one or more remote terminals, possibly remote with respect to the local server, intended to carry out only the simulations and the connected large number of calculations. The local server can be a computer, properly programmed to carry out the training of the machine learning model 4, and achieving the calculation of the model parameters of the model 4 by the remote terminals.

[0035] The machine learning model 4, upon being properly trained, along with the model parameters, is transferred to the second processing unit 33.

[0036] The second processing unit 33 can be implemented in several ways. In fact, the second processing unit 33 can be a microprocessor. Microprocessors are highly versatile and programmable and can be embedded in a wide variety of electronic schemes. The primary advantage of using microprocessors lies in their ability to execute a wide array of instructions and manage multiple tasks simultaneously. They offer high computational power in a compact form factor, making them suitable for integration intodifferent systems. Furthermore, microprocessors are widely supported by a robust ecosystem of software development tools and libraries, facilitating rapid application development and deployment.

[0037] In some other embodiments, the second processing unit 33 can be implemented through Field-Programmable Gate Arrays (FPGAs). FPGAs offer the advantage of reconfigurability, allowing developers to tailor the hardware functionality to specific application needs even after deployment. This flexibility makes FPGAs ideal for applications requiring specialized processing capabilities and adaptability. Additionally, FPGAs can achieve high levels of parallelism, significantly improving the performance for tasks that can be executed concurrently.

[0038] In other embodiments, the second processing unit 33 can be implemented as a personal computer (PC). PCs are well-suited for a wide range of applications. The advantage of implementing the second processing unit 33 in a personal computer is the extensive computational resources available, including advanced graphics processing units (GPUs) and high-speed memory.

[0039] An additional implementation option is the use of Application-Specific Integrated Circuits (ASICs). ASICs are custom-designed for specific tasks, providing optimal performance and efficiency for particular applications. An advantage of ASICs is their ability to deliver high performance with minimal power consumption.

[0040] The second processing unit 33 is connected to the centrifugal compressor 2, to receive the stream of operating data to be processed by the machine learning model 4, as better explained below.

[0041] Centrifugal compressors are dynamic machines that utilize the principle of centrifugal force to increase the pressure and velocity of a gas. The centrifugal compressors are widely used in various industries, including HVAC, power generation, petrochemical, and aerospace, to handle large volumes of gas with minimal maintenance. The basic operation of a centrifugal compressor involves a rotating impeller that imparts kinetic energy to the gas. This energy is then converted into pressure energy as the gas passes through a diffuser and volute casing.

[0042] The impeller consists of a senes of blades mounted on a rotating shaft. As the impeller spins, gas is drawn into the center (eye) of the impeller and expelled outward along the blades by centrifugal force. The gas exits the impeller at high velocity and enters the diffuser, a stationary passage that gradually expands to decrease the gas velocity while increasing its pressure. The gas is then collected in the volute casing, where it is directed to the discharge point.

[0043] In other embodiments, the predicting system 1 can be applied to other equipment, such as single-stage and multi-stage centrifugal compressors, or integrally geared centrifugal compressors, radial and mixed flow centrifugal compressors, mixed flow centrifugal compressors, depending on the application required.

[0044] The operation of any equipment where the predicting system is applied to, can be characterized by a vector or several operating parameters. These operating parameters may include but are not limited to, inlet pressure, inlet temperature, outlet pressure, outlet temperature, mass flow rate, gas composition, and rotational speed.

[0045] The second processing unit 33 is connected to the centrifugal compressor 2, allowing the reception of real-time generated data from the centrifugal compressor 2 to be input into the machine learning model 4. The second processing unit 33 uses this data to predict mass flow rate at surge conditions in the centrifugal compressor 2, and to calculate the relative distance with the current flow rate. Fig. 6 shows an exemplary centrifugal map CM where simulated data and the prediction data are plotted for different operating points.

[0046] The surge prediction system 1 may enable communication between the offline and operating environments or setup, allowing, in some embodiments, the machine learning model 4 to be updated with new operating data collected during the compressor 2 operation. This facilitates continuous improvement of the surge prediction capabilities of the surge prediction system 1.

[0047] In some aspects, the surge prediction system 1 may comprise one or more sensors 22 configured to measure the plurality of performance or operating parameters of the centrifugal compressor 2 or the equipment it is applied to. Such sensors 22 may be those normally implemented in the equipment, or, in the case at issue, in the centrifugalcompressor 2.

[0048] Specifically a data acquisition unit 21 is configured to receive the plurality of operating data in real-time from the above sensors 22, in real-time, then the second processing unit 33 executes a method 100 for predicting surge in the centrifugal compressor 2 (see Figures 2, 3, 4, and 5).

[0049] In some cases, the machine learning model 4 is implemented on a neural network, like a deep neural network (DNN) and / or Gaussian Processes (GP) and / or Support Vector Regression (SVR) and / or Gradient Boosted Trees (GBDT) and / or other tree-based methods.

[0050] Also, the machine learning model 4 implemented in the neural network may be the Polynomial Ridge Regression. This type of model may be particularly suitable for predicting surge events due to its ability to handle multi-dimensional input data and its robustness against overfitting. The Polynomial Ridge Regression model may use the performance parameters as input variables to predict the surge distance, which is calculated as the difference between the current mass flow rate and the predicted mass flow rate at surge.

[0051] Other machine learning models can be implemented in the neural network 4, without departing from the scope of the present disclosure. In particular, the models of and / or Gaussian Process Regression and / or Deep Neural Network and / or Gradient Boosted Tree Regression can be implemented.

[0052] The surge prediction system 1 comprises a monitoring system that uses experimental signals acquired from the centrifugal compressor 2 as input for the surrogate models. This monitoring system provides real-time information on surge distance during tests, allowing test engineers to work in a more efficient way.

[0053] The second processing unit 33 can also implement or comprise a monitoring system to trigger an alarm 35 when the mass flow rate is below the surge limit predicted by the model or different from a determined threshold, prompting the test engineer to perform corrective actions to limit the risk of incurring a surge event. This realtime monitoring and alert system may significantly enhance the safety and efficiencyof operating the centrifugal compressor 2.

[0054] The alarm 35 can be physical, like an alarm light or a horn, or virtual, as an alert generated on the display 331 of the second processing unit 33.

[0055] Still referring to Fig. 1, the surge prediction system 1 may be designed to facilitate continuous improvement of the surge prediction capabilities. This may be achieved through communication between the offline and environment environments, allowing the machine learning model 4 to be updated with new operational data collected during the regular operation of the centrifugal compressor 2.

[0056] The data acquisition unit 21 may collect real-time performance parameters from the centrifugal compressor 2 during operation and transmit this data to the second processing unit 33. The second processing unit 33 may then use this real-time data as input for the machine learning model 4, allowing the model to adapt to changing operational conditions and improve its surge prediction accuracy over time.

[0057] The operation of the prediction system 1 is as follows, with particular reference to Figures 2, 3, and 4.

[0058] The second processing unit 33 processes the real-time operating data acquired by the sensors 22 executing a method 100. Specifically, referring to Figures 2, 3, and 4, the flowcharts of the method 100 are illustrated, and comprise the steps of receiving a plurality of real-time operating data acquired by the sensors 22 of the centrifugal compressor 2, to be used as input for the operation of the machine learning model 4 implemented in the second processing unit 33, to predict the surge distance.

[0059] As mentioned, the machine learning model 4 is trained on simulated data generated by the first processing unit 31 previously. The second processing unit 33, by the trained machine learning model 4, compares the predicted surge distance to a predetermined threshold, and generates an alert signal by the alarm 35 when the predicted surge distance is below or above the predetermined threshold.

[0060] More specifically, the method 100 for predicting surge in an equipment, such as a centrifugal compressor 2 begins with step 110, where the second processing unit 33 receives the operating data of the equipment, namely, in the embodiment at issue,the centrifugal compressor 2.

[0061] Following step 110, the method 100 may proceed to a following step (not shown in the figure), where the second processing unit 33 corrects the received performance parameters based on certain factors, such as gas composition and machine component sizing, to obtain adjusted operating data.

[0062] In step 120, the second processing unit 33 predicts a surge distance measure using the machine learning model 4 trained on the simulated data through the first processing unit 31.

[0063] The surge distance may be calculated as a difference between a current mass flow rate and a predicted mass flow rate at surge. The machine learning model 4 uses the received performance parameters as input variables to predict the surge distance.

[0064] In step 130, the first processing unit 33 compares the predicted surge distance to a predetermined threshold. The predetermined threshold may be set based on certain criteria, such as safety margins or operational limits of the equipment 2. If the predicted surge distance is below (or above, depending on the settings) the predetermined threshold, it may indicate a potential risk of surge occurrence in the centrifugal compressor 2.

[0065] In step 140, the second processing unit 33 may generate an alert signal by the alarm 34, when the predicted surge distance is below the predetermined threshold. The alert signal may be used to notify a test engineer or an automatic control system of the potential risk of surge occurrence, prompting them to take corrective actions to prevent the surge event. This real-time alert system improves safety and efficiency of operating the equipment 2.

[0066] Referring to Figures 3 and 4, additional steps in the surge prediction process are illustrated. In some aspects, the method 100 may further include a step 150 of predicting outlet pressure and temperature using the same or another machine learning model 4 trained on simulated data. The machine learning model 4 (or another machine learning model) may use the received performance parameters as input variables to make the prediction, in this case.

[0067] In some cases, the method 100 may also comprise a step 160 of verifying the accuracy of the machine learning model 4. This verification step 160 may involve comparing predicted outlet pressure and temperature to measured outlet pressure and temperature. The comparison may be performed by the second processing unit 33, and the results may be used to assess the accuracy of the one or more machine learning models 4. If the predicted values closely match the measured values, it would indicate that the machine learning models 4 and the original physical simulation system are accurately predicting the performance parameters of the centrifugal compressor 2.

[0068] In some aspects, the method 100 may further comprise a step 170 of storing the received performance parameters and predicted surge distances in the data storing unit 32. The data storing unit 32 may be connected to the first processing unit 31 and may be configured to store a variety of data, including but not limited to, the received performance parameters of the machine learning model 4, the predicted surge distances, and the results of the verification step 160.

[0069] The stored data may be used for various purposes, such as retraining the machine learning model 4, analyzing the performance of the centrifugal compressor 2, or identifying trends in the operating conditions of the centrifugal compressor 2.

[0070] In some cases, the method 100 may also include a step 180 of retraining the machine learning models 4 using the stored data to improve prediction accuracy over time. The retraining step 180 may be performed by the first processing unit 31 and may involve updating the weights and biases of the machine learning models 4, namely the based on the stored data.

[0071] This continuous learning approach allows the surge prediction system 1 to adapt to changing operating conditions and improve its surge prediction accuracy over time. The retraining step 180 may be performed periodically or whenever a significant change in the operating conditions of the centrifugal compressor 2 is detected.

[0072] In the offline environment, the method 100 described herein involves a training procedure 190 for optimizing the performance of a machine learning model 4. The training procedure 190 comprises two primary steps: generating 191 simulated data and training 192 the machine learning model 4.

[0073] The generation of simulated data 191 is executed by simulation software that leverages analytical empirical formulas, maps, and numerical models. These tools are capable of simulating the effects of varying gas compositions and the sizing of machine components on the equipment 2. Analytical empirical formulas are mathematical representations derived from experimental data, which provide approximations of physical phenomena. Maps in this context refer to datasets containing input and output variables that can be interpolated to extract the relationships between different variables affecting the equipment’s performance. Numerical models are computational algorithms that solve equations representing physical systems, allowing for the prediction of how different variables interact and affect outcomes. The simulation software uses these components to create a dataset that reflects the potential real-world scenarios the equipment 2 might encounter. This simulated data are used for training the machine learning model 4 as it encompasses a wide range of possible conditions and configurations, ensuring the model can generalize well to actual operating environments. The choice of the combinations of operating conditions and gas composition that need to be simulated may be defined a priori following specifications or design of experiment methodologies, as well as iteratively via a procedure of active learning. This procedure is commonly known by experts in the field and consists of collect small batches of samples and let the machine learning model inherent uncertainty measure identify the next points that need to be simulated in order to iteratively increase its predictive performance in a more data-efficient way.

[0074] In the training step 192, the at least one machine learning model 4 is trained using the simulated data generated in the previous step. This involves calculating the model parameters essential for operating the machine learning model 4. The training process adjusts these parameters to minimize the error between the model’ s predictions and the known outcomes from the simulated data.

[0075] The resulting trained machine learning models 4 are then capable of making predictions about the performance of the equipment 2 under various conditions.

[0076] Potentially, the processing means 31, 33 comprise a first processing unit 31 and a storing unit 32, wherein the storing unit 32 is connected to the first processing unit 31 and the second processing unit 33 is connected to or installed in the equipment2. The first processing unit 31 can executes the training procedure 190 and the second processing unit 33 can executes the other steps of the method 100.

[0077] Referring to Fig. 6, a graph is depicted showing the compressor map, which P2 discloses the relationship between flow rate QI and the pressure ratio — namely theratio of the output pressure P2over input pressure P1for a centrifugal compressor 2.

[0078] The graph comprises three different data series: a predicted curve represented by a dashed line, an SCL (Surge Control Line) curve represented by a solid line with circular markers, and target points represented by open circular markers. The X-axisP2 may represent QI values, namely the flow rate, while the Y-axis reports — values.

[0079] Multiple curves are shown for different rotational speed, with each curve displaying an upward trend followed by a downward trend as QI increases. The predicted curves closely follow the target points, indicating good agreement between the prediction model and the actual data. This close agreement suggests that the machine learning model is accurately predicting the performance parameters of the centrifugal compressor 2, thereby enhancing the reliability of the surge prediction system 1.

[0080] In some cases, the predicted curve may be generated by the machine learning model 4 using the received model parameters as input to operate the one or more machine learning models 4. The SCL curve represents the surge control line, which is a boundary line that separates stable operation from potentially unstable operation in the performance map of the centrifugal compressor 2. The target points represent performance parameters obtained with detailed simulations of the operation of the centrifugal compressor 2.

[0081] In other aspects, the graph may be used to monitor the performance of the centrifugal compressor 2 in real-time. The position of the predicted curve relative to the SCL curve and the target points may provide information about the current operating conditions of the centrifugal compressor 2 and the risk of surge occurrence. If the predicted curve approaches or crosses the SCL curve, it may indicate a potential risk of surge occurrence, prompting the surge prediction system 1 to generate an alert signal by the alarm 35.

[0082] In some cases, the graph of Fig. 6 may also be used to assess the accuracy of the one or more machine learning models 4. The closeness of the predicted curve to the target points may indicate the accuracy of the machine learning models in predicting the performance parameters of the centrifugal compressor 2. This information may be used to continuously improve the one or more machine learning model 4 through retraining with newly acquired operational data.

[0083] The surge prediction system and method described herein have wide applicability in industrial settings where centrifugal compressors are utilized. This invention is particularly relevant in industries such as oil and gas, chemical processing, power generation, and refrigeration, where centrifugal compressors play a critical role in various processes. The system's ability to predict surge events in real-time and provide early warnings can significantly improve the safety, efficiency, and reliability of compressor operations across these industries. While the disclosure primarily focuses on centrifugal compressors, the principles and techniques described may be adapted for use with other types of compressors or turbomachinery where surge prediction is beneficial. The adaptive nature of the machine learning models allows the system to be applied to a diverse range of compressor designs and operating conditions, making it versatile for different industrial applications. Furthermore, the system's capability to continuously improve its prediction accuracy over time makes it particularly valuable for long-term industrial use, potentially leading to optimized compressor performance and reduced maintenance costs across various industrial sectors.

[0084] While aspects of the invention have been described in terms of various specific embodiments, it will be apparent to those of ordinary skill in the art that many modifications, changes, and omissions are possible without departing form the spirt and scope of the claims. In addition, unless specified otherwise herein, the order or sequence of any process or method steps may be varied or re-sequenced according to alternative embodiments.ADVANTAGES

[0085] The surge prediction system for centrifugal compressors of the present disclosure offers several significant technical advantages. In fact, the system's use of machine learning models trained on simulated data allows for surge prediction withoutrequiring extensive historical operational data. This is particularly useful for new or modified compressor installations where such historical data may not be available. The system's ability to predict surge distance in real-time provides an approach to surge prevention, allowing operators to take preventive actions before critical conditions are reached, thereby enhancing operational safety and efficiency.

[0086] Advantageously, using simulated data instead of experimental data for the control system enables the technology to be really useful. For example, if a user is running a machine for the first time, the user does not have any experimental data; this allows to predict the behavior based on design features, that is especially useful for equipment that are assembled once only and not in series.

[0087] In addition, using simulated data instead of only experimental data allows to reach the following advantages: start before the real machine is installed (data not yet collected / available for tested machine), collect larger variations in input variables that experimentally could not be reached, and evaluate a variable that is very complex to evaluate experimentally (surge point), which could never be available from pure experimental data

[0088] To improve efficiency and quality by using simulated data it may be necessary to create a representative and validated model and / or set up a complex model monitoring workflow for the differences between experiments and simulation results.

[0089] As another advantage, the system's adaptability to different gas compositions and machine component sizing addresses the complexity of compressor behavior under varying conditions. This makes the system applicable across a wide range of industrial applications and compressor designs. The integration of multiple machine learning models for predicting several parameters (surge distance, mass flow rate at choke, outlet pressure, and temperature) enables a comprehensive assessment of compressor performance, enhancing the overall reliability of the predictions.

[0090] The self-improving feature ensures that the system becomes increasingly tailored to the specific characteristics of the compressor, it monitors over time, potentially leading to more efficient compressor operation and reduced probability of surge events.

[0091] Additionally, the real-time processing capabilities of the system allow for immediate response to changing operational conditions, to prevent surge in dynamic industrial environments.

[0092] Reference has been made in detail to embodiments of the disclosure, one or more examples of which are illustrated in the drawings. Each example is provided by way of explanation of the disclosure, not limitation of the disclosure. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made in the present disclosure without departing from the scope or spirit of the disclosure. Reference throughout the specification to "one embodiment" or "an embodiment" or “some embodiments” means that the particular feature, structure or characteristic described in connection with an embodiment is included in at least one embodiment of the subject matter disclosed. Thus, the appearance of the phrase "in one embodiment" or "in an embodiment" or "in some embodiments" in various places throughout the specification is not necessarily referring to the same embodiment(s). Further, the particular features, structures or characteristics may be combined in any suitable manner in one or more embodiments.

[0093] When elements of various embodiments are introduced, the articles “a”, “an”, “the”, and “said” are intended to mean that there are one or more of the elements. The terms “comprising”, “including”, and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements.

[0094] The subject matter described herein can be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structural means disclosed in this specification and structural equivalents thereof, or in combinations of them. The subject matter described herein can be implemented as one or more computer program products, such as one or more computer programs tangibly embodied in an information carrier (e.g., in a machine-readable storage device), or embodied in a propagated signal, for execution by, or to control the operation of, data processing apparatus (e.g., a programmable processor, a computer, or multiple computers). A computer program (also known as a program, software, software application, or code) can be written in any form of programming language, including compiled or interpreted languages, and it can be deployed in any form, including as astand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file. A program can be stored in a portion of a file that holds other programs or data, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers at one site or distributed across multiple sites and interconnected by a communication network.

[0095] The processes and logic flows described in this specification, including the method steps of the subject matter described herein, can be performed by one or more programmable processors executing one or more computer programs to perform functions of the subject matter described herein by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus of the subject matter described herein can be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).

[0096] Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from a read-only memory, or a random access memory, or both. The essential elements of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, (e.g., EPROM, EEPROM, and flash memory devices); magnetic disks, (e.g., internal hard disks or removable disks); magneto-optical disks; and optical disks (e.g., CD and DVD disks). The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0097] To provide for interaction with a user, the subject matter described herein canbe implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, (e.g., a mouse or a trackball), by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well. For example, feedback provided to the user can be any form of sensory feedback, (e.g., visual feedback, auditory feedback, or tactile feedback), and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0098] The techniques described herein can be implemented using one or more modules. As used herein, the term “module” refers to computing software, firmware, hardware, and / or various combinations thereof. At a minimum, however, modules are not to be interpreted as software that is not implemented on hardware, firmware, or recorded on a non-transitory processor readable recordable storage medium (i.e., modules are not software per se). Indeed “module” is to be interpreted to always include at least some physical, non-transitory hardware such as a part of a processor or computer. Two different modules can share the same physical hardware (e.g., two different modules can use the same processor and network interface). The modules described herein can be combined, integrated, separated, and / or duplicated to support various applications. Also, a function described herein as being performed at a particular module can be performed at one or more other modules and / or by one or more other devices instead of or in addition to the function performed at the particular module. Further, the modules can be implemented across multiple devices and / or other components local or remote to one another. Additionally, the modules can be moved from one device and added to another device, and / or can be included in both devices.

[0099] The subj ect matter described herein can be implemented in a computing system that includes a back-end component (e.g., a data server), a middleware component (e.g., an application server), or a front-end component (e.g., a client computer having a graphical user interface or a web browser through which a user can interact with an implementation of the subject matter described herein), or any combination of such back-end, middleware, and front-end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a com-munication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet.

Claims

Surge Prediction Method and System for Centrifugal CompressorsCLAIMS1. A computer-implemented method (100) for predicting surge in an equipment (2), wherein the method (100) comprises the steps of: receiving (110), by processing means (31, 33), a plurality of operating data of an equipment (2); predicting (120), by the processing means (31, 33), a surge distance using at least one machine learning model (4), wherein the surge distance is calculated as a difference between a current flow rate and a predicted flow rate at surge; comparing (130), by the processing means (31, 33), the predicted surge distance to a predetermined threshold; generating (140), by the processing means (31, 33), an alert signal when the predicted surge distance is below the predetermined threshold; and wherein the at least one machine learning model (4) is trained by training procedure (190) based on simulated data.

2. The method (100) of claim 1, wherein the receiving (110) step comprises the sub-step of adjusting the received plurality of operating data of an equipment (2), to obtain adjusted operating data.

3. The method (100) of any one of the preceding claims, wherein the plurality of operating data of the equipment (2) comprises at least one of the following: inlet pressure; inlet temperature; outlet pressure; outlet temperature; mass / volumetric flow rate; gas composition; and rotational speed.

4. The method (100) of any one of the preceding claims, further comprising the step of further predicting (150), by the processing means (31, 33), outlet pressure and temperature by the at least one machine learning model (4).

5. The method (100) of any one of the preceding claims, further comprising the step of verifying (160) the accuracy of the at least one machine learning model (4) by comparing predicted outlet pressure and temperature to measured outlet pressure and temperature.

6. The method of any one of the preceding claims, wherein the training procedure (190) comprises the steps of: generating (191) the simulated data by a simulation software; and training (192) the at least one machine learning model (4), calculating the model parameters for operating the machine learning model (4).

7. The method (100) of any one of the preceding claims, further comprising the steps of: storing (170) the received performance parameters and predicted surge distances in a database; and retraining (180) the at least one machine learning models (4) using the stored data to improve prediction accuracy over time.

8. The method (100) of any one of the preceding claims, wherein the machine learning model (3) is the Polynomial Ridge Regression and / or Gaussian Process Regression and / or Deep Neural Network and / or Gradient Boosted Tree Regression.

9. The method (100) of any one of the preceding claims, wherein the machine learning model is based on a deep neural network (DNN) and / or Gaussian Processes (GP) and / or Support Vector Regression (SVR) and / or Gradient Boosted Trees (GBDT) and / or other tree-based methods (3).

10. The method (100) of any one of the preceding claims,wherein the equipment (2) comprises one or more sensors (22) for acquiring operating data of the equipment (2), and wherein the processing means (31, 33) are connected to the one or more sensors (22), to receive the operating data acquired.

11. The method (100) of any one of the preceding claims, wherein the processing means (31, 33) comprise: first processing unit (31); a storing unit (32), connected to the first processing unit (31); and second processing unit (33), connected to, or installed in the equipment (2); wherein the first processing unit (31) executes the training procedure (190); and wherein the second processing unit (33) executes the other steps of the method (100).

12. The method (100) of any one of the preceding claims, wherein the equipment is a centrifugal compressor (2).

13. A system (1) for predicting surge in an equipment (2), characterized in that the system (1) comprises: one or more sensors (22) configured to measure one or more operating data of the equipment (2); processing means (31, 33), connected to the sensors (22); and a storing unit (32) that, when executed by the processing means (31, 33), cause the system (1) to execute the method (100) according to any one of the preceding claims.

14. The system (1) of the preceding claim, wherein the equipment is a centrifugal compressor (2).

15. A non-transitory computer-readable storage medium storing instructions that, when executed by the processing means (31, 33), cause the processing means (31, 35) to perform a method (100) for predicting surge in a equipment (2)according to any one of claims 1-12.

Images