Online classification method, device, equipment and medium for thermal quench mode of radio frequency superconducting cavity based on machine learning

Abstract

The present invention discloses a method, device, equipment and medium for online classification of thermal quench modes of radio frequency superconducting cavities based on machine learning, including: obtaining the cavity pressure of the radio frequency superconducting cavity; V c and forward voltage V f According to the cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ; Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ; The extracted features Q id , E id and N soft The fault diagnosis model is input to realize fault classification, and the faults include non-thermal quench faults and thermal quench faults. Thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench, and soft quench. Therefore, the present invention can realize automatic online classification of thermal quench modes and provide high-accuracy fault information feedback in seconds for accelerator operators.

Description

Technical Field

[0001] The present invention relates to a method, device, equipment and medium for online classification of radio frequency superconducting cavity thermal quench modes based on machine learning, and relates to the field of particle accelerators. Background Art

[0002] Radio frequency superconducting cavities (hereinafter referred to as superconducting cavities) are very suitable for accelerating high average current beams in continuous wave or long pulse mode due to their low loss and high gradient characteristics. In recent years, many major accelerator projects at home and abroad have chosen superconducting technology as the preferred solution. However, the operating bandwidth of superconducting cavities is usually limited to tens to hundreds of hertz. The extremely narrow bandwidth makes it easy to detune when disturbed, resulting in failures. Therefore, real-time monitoring of digital low-level feedback control systems is required to maintain stability.

[0003] Thermal quench is a serious fault of superconducting cavity, which means that the heat generated on the cavity surface cannot be dissipated in time, causing the temperature to exceed the critical point. The thermal quench fault of a single cavity can spread rapidly, causing collective failures and seriously affecting beam operation. Thermal quench will increase energy consumption and make the cavity's no-load quality factor ( Q 0) and loaded quality factor ( Q L ) dropped rapidly, eventually triggering a machine failure.

[0004] In recent years, advanced accelerator laboratories, represented by the European Synchrotron Radiation Laboratory (DESY), have been Q L The thermal quench event was successfully identified by the change of the thermal quench event. However, thermal quench often exhibits different behavior modes, corresponding to different physical mechanisms. According to the operation experience of the superconducting linear accelerator CAFE2 device of the Institute of Modern Physics, thermal quench events can be divided into three different modes. Conventional thermal quench is usually sporadic, and before the thermal quench occurs (i.e. Q L Before significant changes), cavity pressure and cavity incident, reflected signal, dynamic cavity detuning and dynamic Q L The signal usually does not have a precursor. Another type of thermal quench event is also called electrical quench-induced thermal quench. The main feature is that the thermal quench is usually preceded by an electrical quench. The electrical quench itself is different from the thermal quench. It is an independent type of superconducting cavity failure caused by field emission (FE). A small number of FE electrons in the electrical quench event will bombard the cavity wall, causing local overheating and eventually evolving into a thermal quench. Before the third type of thermal quench event occurs, the superconducting cavity Q L There will be a period of regular passivity. Q LThe drop in the temperature is much smaller than that of conventional thermal quench, which is also called soft quench. In short, different thermal quench events may correspond to different physical mechanisms, requiring on-site personnel to take different treatment measures.

[0005] In summary, although the existing technology can identify thermal quench events, given that different thermal quench modes require different processing strategies, it is urgent to develop a method that can distinguish thermal quench modes online. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, in view of the above problems, the purpose of the present invention is to provide a method, device, equipment and medium for online classification of thermal quench modes of radio frequency superconducting cavities based on machine learning, which can realize automatic online classification of thermal quench modes and provide high-accuracy fault information feedback in seconds for accelerator operators.

[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is:

[0008] In a first aspect, the present invention provides an online classification method for thermal quench modes of radio frequency superconducting cavities based on machine learning, the method comprising:

[0009] Obtaining the cavity pressure of a radio frequency superconducting cavity V c and forward voltage V f ;

[0010] According to cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ;

[0011] Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ;

[0012] The features to be extracted Q id , E id and N softThe fault diagnosis model is input to realize fault classification. The faults include non-thermal quench faults and thermal quench faults. Thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench and soft quench.

[0013] Further, Q id It is used to distinguish thermal quench events from non-thermal quench events. The calculation formula is:

[0014]

[0015] In the formula, e =| r c ─ r |, T Q is the integration time.

[0016] Further, E id Used to distinguish electrical quench events, E id The calculation formula is:

[0017]

[0018] In the formula, Δ r [n] is the differential signal of the cavity pressure amplitude, .

[0019] Further, N soft Used to distinguish soft quench events, N soft Characteristic statistical deviation signal e The number of times the lower and upper boundaries are crossed.

[0020] Furthermore, the training of the fault diagnosis model uses the random forest algorithm to build the fault diagnosis model, and the grid search method is used to optimize the model hyperparameters.

[0021] Furthermore, before the training of the fault diagnosis model, the process of manually labeling the thermal quench events is also included, including:

[0022] Based on dynamic loaded quality factor Q L Distinguish thermal quench and non-thermal quench events and manually mark them;

[0023] When a thermal quench event occurs, the thermal quench mode is manually marked: conventional thermal quench has no precursors before the occurrence of thermal quench, and the dynamic load quality factor Q LThere is no obvious abnormality in the curve before the thermal quench; the electrical quench-induced thermal quench event is observed before the thermal quench, and the cavity pressure amplitude drops sharply due to the electrical quench; the soft quench event will cause the dynamic load quality factor Q L The curve shows regular fluctuations.

[0024] Furthermore, the dynamic loaded quality factor Q L The calculation formula is:

[0025]

[0026] In the formula, ω 0 is the cavity resonance frequency, ω h is the cavity half bandwidth.

[0027] In a second aspect, the present invention further provides an online classification device for thermal quench modes of radio frequency superconducting cavities based on machine learning, comprising:

[0028] The voltage signal module is configured to obtain the cavity pressure of the radio frequency superconducting cavity. V c and forward voltage V f ;

[0029] The signal calculation module is configured to calculate the cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ;

[0030] Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ;

[0031] The features to be extracted Q id , E id and N soft The fault diagnosis model is input to realize fault classification. The faults include non-thermal quench faults and thermal quench faults. Thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench and soft quench.

[0032] In a third aspect, the present invention provides an electronic device comprising: at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute any one of the methods described.

[0033] In a fourth aspect, the present invention provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include computer instructions, and the computer instructions are used to enable a computer to execute any one of the methods described.

[0034] The present invention adopts the above technical solution, and has the following characteristics:

[0035] 1. The present invention can realize automatic online classification of thermal quench modes and provide accelerator operators with high-accuracy fault information feedback in seconds.

[0036] 2. The data used in the present invention are all based on the fault data of the accelerator operation. No special measurement experiments are required. The online deployment and debugging of the machine learning model can be run in parallel with the beam flow without taking up additional machine running time.

[0037] 3. The present invention completes feature extraction and machine learning classification by reading RF fault signals. Among them, the acquisition, down-conversion and baseband signal demodulation of RF signals can all be completed by a general digital low-level system, and data preprocessing, feature extraction algorithms and machine learning models can be deployed on a general host computer. The above applications do not require additional hardware equipment.

[0038] In summary, the present invention can be widely used in particle accelerators. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Throughout the accompanying drawings, the same reference numerals are used to represent the same components. In the accompanying drawings:

[0040] Figure 1 FIG. 4 is a block diagram of a device for classifying thermal quench modes according to an embodiment of the present invention.

[0041] Figure 2 The figure is a flow chart of a method according to an embodiment of the present invention.

[0042] Figure 3: is a schematic diagram of the signal normalization principle of an embodiment of the present invention, wherein (a) is a waveform diagram of the amplitude (upper) and phase (lower) of the cavity pressure and the forward voltage before signal normalization; (b) is a waveform diagram of the amplitude (upper) and phase (lower) of the cavity pressure and the forward voltage after signal normalization; the steady-state average value of the amplitude of the cavity pressure is 1, and the steady-state average value of the phase of the cavity pressure is 0 degrees.

[0043] Figure 4 Figure 1 is a comparison diagram of signals before and after filtering using different digital filters according to the embodiment of the present invention, wherein (a) and (b) are amplitude comparisons before and after filtering, respectively; (c) is a comparison diagram of the amplitude before and after filtering according to the amplitude before and after filtering. V c, org and after filtering V c Computational Dynamics Q L The waveform, for comparison, also shows the static Q L (Right now Q L, ss ) results.

[0044] Figure 5 The waveform diagrams of three typical thermal quench events and non-thermal quench faults involved in the embodiments of the present invention are shown in Figures (a) to (d), respectively: conventional thermal quench, thermal quench caused by electrical quench, soft quench and non-thermal quench.

[0045] Figure 6 for Figure 5 The dynamics corresponding to the four events in Q L and dynamic detuning Δ f The measured results.

[0046] Figure 7 The characteristics of thermal quench and non-thermal quench in the embodiment of the present invention are shown in FIG. Q id comparison.

[0047] Figure 8 The features of the embodiments of the present invention are N soft Calculation principle diagram.

[0048] Fig. 9 for Figure 5 The cavity pressure amplitude measurements for the four events shown are r , predicted value of cavity pressure amplitude r c And the difference between the two e ( e =| r c - r |).

[0049] Fig.10 Three features in the embodiment of the present invention Q id , E id , N soft The distribution of conventional thermal quench (Q), electrical quench-induced thermal quench (EQ), soft quench (SQ) and non-thermal quench (NQ) events, where (a), (b) and (c) are the three characteristics corresponding to different situations;

[0050] Fig.11 This is a principal component analysis (PCA) diagram of four events, namely, conventional thermal quench (Q), electrical quench-induced thermal quench (EQ), soft quench (SQ) and non-thermal quench (NQ), observed on the CAFE2 device of the Institute of Modern Physics in an embodiment of the present invention.

[0051] Fig.12 The random forest model is used to obtain the confusion matrix and F1-Score matrix results in the embodiment of the present invention.

[0052] Fig.13 FIG. 4 is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0054] Although the terms first, second, third, etc. can be used in the text to describe multiple elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms can only be used to distinguish an element, component, region, layer or section from another region, layer or section. Unless the context clearly indicates, terms such as "first", "second" and other numerical terms do not imply order or sequence when used in the text. Therefore, the first element, component, region, layer or section discussed below can be referred to as the second element, component, region, layer or section without departing from the teaching of the example embodiments.

[0055] For ease of description, spatially relative terms may be used herein to describe the relationship of one element or feature relative to another element or feature as shown in the figures, such as "inside", "outside", "inner side", "outer side", "below", "above", etc. Such spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation depicted in the figures.

[0056] The existing technology can only automatically identify thermal quench events in superconducting cavity faults, and the specific thermal quench mode identification still requires manual intervention. The online classification method, device, equipment and medium of the radio frequency superconducting cavity thermal quench mode based on machine learning provided by the present invention include: obtaining the cavity pressure of the radio frequency superconducting cavity; V c and forward voltage V f According to the cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ; Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ; The extracted features Q id , E id and N soft The machine learning model is input to realize fault classification. Faults include non-thermal quench faults and thermal quench faults. Thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench, and soft quench. Therefore, the present invention can realize automatic online classification of thermal quench modes and provide high-accuracy fault information feedback in seconds for accelerator operators.

[0057] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0058] This embodiment takes the CAFE2 device as an example to provide corresponding data acquisition and describe the deployment of the machine learning model training process to distinguish the thermal quench mode of the superconducting cavity online, mainly including the following aspects: data preprocessing, thermal quench data marking, feature acquisition, and machine learning algorithm.

[0059] Embodiment 1: The online classification method of radio frequency superconducting cavity thermal quench mode based on machine learning provided in this embodiment includes:

[0060] S1. Measurement of voltage signal.

[0061] In this embodiment, Figure 1 As shown, the measurement of the voltage signal is mainly realized by a digital low-level system 1, a solid-state power source 2, a directional coupler 3, an input coupler 4, a radio frequency cavity 5, a signal extraction coupler 6, a down-conversion module 7 and a host computer 8. Among them, the output end of the digital low-level system 1 is connected to the input end of the solid-state power source 2, the output end of the solid-state power source 2 is connected to the input end of the directional coupler 3, the output end (through port) of the directional coupler 3 is fed into the radio frequency cavity 5 through the input coupler 4, and the directional coupler 3 is used to extract the cavity incident signal of the radio frequency cavity 5. P f and cavity reflection signal P r The signal extraction coupler 6 is connected to the RF cavity 5 to extract the cavity sampling signal P t , cavity incident signal P f , cavity reflection signal P r and cavity sampling signal P t After down-conversion by the down-conversion module 7, the cavity sampling signal is sent to the digital low-level system 1. P t , cavity incident signal P f and cavity reflection signal P r Down-converted to an intermediate frequency signal, the digital signal processing module of the digital low-level system 1 demodulates the intermediate frequency signal through I / Q to obtain three groups of original digital baseband signals, namely: the cavity pressure of the RF cavity V c * , forward voltage V f * and reverse voltage V r *, where the intermediate frequency signal refers to the signal after down-conversion. Whether in the communication field or the RF low-level field, it is not suitable to directly process the RF signal for the following reasons: 1) The limitation of the working frequency band of RF devices such as amplifiers; 2) The limitation of the ADC band, that is, the ADC cannot sample the RF signal with a higher frequency, but can only sample the intermediate frequency signal with a relatively low frequency. This is also the main reason why the RF signal is first down-converted to the intermediate frequency. The three groups of baseband signals are the cavity pressure of the RF cavity. V c * , forward voltage V f * and reverse voltage V r * It will be sent to the host computer 8 for subsequent processing.

[0062] Furthermore, the voltage involved in this embodiment is, for example V c * , V f * , V r etc. are all in complex form, which can be expressed in the form of amplitude and phase, for example: V =| V | e j∠V , where | V | is the amplitude or modulus, ∠ V is the phase or argument; it can also be expressed as the real part plus the imaginary part, for example: V = V I + JbXj Q ,in, V I is the real part or the in-phase component, V Q is the imaginary part or called the orthogonal component.

[0063] Furthermore, in this embodiment P t Represents a radio frequency signal, such as a 162.5 MHz radio frequency signal. V c It means that the digitized baseband signal has no RF frequency components, which are removed through analog and digital down-conversion.

[0064] S2, based on the original cavity pressure V c * and forward voltage V f *The cavity pressure is obtained by calibration and normalization V c and forward voltage V f .

[0065] In this embodiment, the original cavity pressure V c * and forward voltage V f * You can use the formula: V f, cali = XV f * Get the forward voltage after calibration V f, cali , where the correction factor X is plural, X It is mainly used to compensate for the influence of RF signal transmission cables and high-power RF devices on the measurement results. X The existing technology can be used to solve it, which can be expressed as amplitude and phase (i.e. X =| X | e j∠X ) can also be expressed as the real part plus the imaginary part (i.e. X = X I + JXj Q ) form, which is not limited here. Note: The cavity pressure signal is usually used as a reference signal and does not need to be calibrated. The complex multiplication operation can be performed as follows:

[0066]

[0067] In the formula, V f, caliI and V f, caliQ They are V f, cali The real and imaginary parts of V fI * and V fQ * They are V f * The real and imaginary parts of X I and X Q The calibration coefficients are X The real and imaginary parts of .

[0068] Furthermore, if Figure 3 (a) shows the calculation of the original cavity pressure V c * The signal runs at the steady-state signal average value when no electromechanical coupling fault occurs to obtain the normalization coefficient V c, ss * , and use V c, org = V c * / V c, ss * Normalized cavity pressure V c, org Similarly, to ensure signal consistency, the forward voltage after calibration must be V f, cali Perform the exact same scaling to get the normalized forward voltage V f ,Right now V f = V f, cali / V c, ss * .

[0069] Furthermore, in the normalization step, the average value of the cavity pressure during the steady-state time of the superconducting cavity can be calculated as the normalization coefficient V c, ss * Generally, stable data 200 milliseconds before the fault occurs can be selected. For example Figure 3 The superconducting cavity oscillation in (a) occurs at time 0, and the average can be calculated based on the data of -400 milliseconds and -200 milliseconds. Figure 3 (b) shows that after normalization V c The amplitude of the steady-state average is 1 and the phase is 0 degrees. Taking this as an example, the calculation process of finding the average value can be directly called by Python numpy.mean function.

[0070] S3. Establish cavity differential equation and solve dynamic Q L .

[0071] In this embodiment, the cavity differential equation is established to solve the dynamic Q L ,include:

[0072] S31, filtering the cavity pressure before differentiating it helps improve dynamics Q LThe calculation accuracy of the normalized cavity pressure V c, org Filter to get the filtered cavity pressure V c .

[0073] In order to avoid the phase lag caused by the filtering process, this embodiment may select a zero-phase filter (the scipy.signal.filtfilt function in Python). Figure 4 Comparing the 0-phase filter with the traditional FIR (Finite Impulse Response) filter, it does not cause signal lag. After filtering by the 0-phase filter, the cavity pressure V c The high frequency noise in the Q L The calculation accuracy of .

[0074] Furthermore, the parameters of the filtfilt function are selected as follows: the passband frequency is 0.05 times the Nyquist frequency, the stopband frequency is 0.1 times the Nyquist frequency, the stopband attenuation is 75 decibels, the design method uses the Kessel window method, the filter order N and the window parameter beta are calculated using scipy.signal. kaiserord based on the above three parameters, and the final filter order N is 187 and beta is 7.30626. This is an example, not limited to this.

[0075] S32, based on the filtered cavity pressure V c and forward voltage V f Establish the cavity differential equation:

[0076]

[0077] in, Cavity pressure V c The differential of . ω 0 is the cavity resonance frequency, ω h is the cavity half bandwidth, Δ ω is the dynamic cavity detuning, ω 0. ω h and Δω The units are radians per second. f 0. f h With Δ f It is also used to express the resonant frequency, half bandwidth, and detuning of a cavity, but they are not related to ω 0. ω h and Δω The units are different. f 0. f h With Δ f The units of are all Hertz. The corresponding relationships of the above physical quantities are as follows:

[0078]

[0079] External quality factor Q E Generally, it does not change, and it is related to the unloaded quality factor of the cavity. Q 0. Load quality factor Q L The relationship between them is as follows:

[0080]

[0081] When the superconducting cavity is operating stably (no fault occurs), the no-load quality factor Q 0 is much larger than the external quality factor Q E , so 1 / Q L ≈1 / Q E If the loaded quality factor of the superconducting cavity during stable operation is defined as Q L,ss ,but Q L,ss ≈ Q E Established. With external quality factor Q E Same, loaded quality factor Q L,ss Generally, it does not change. Note that for simplicity, this example assumes that the beam current intensity can be ignored, which is applicable to most heavy ion accelerators (i.e., machines with weak beam current load). When there is a beam current, it is only necessary to calculate the cavity dynamic load quality factor. Q L When the beam current is taken into account, the other steps are the same.

[0082] S33, press the cavity V c , forward voltage V f According to the real and imaginary parts, it is decomposed into I + Jb " is in the following form:

[0083]

[0084] Defining parameters ω E = ω 0 / Q E And into the above cavity differential equation:

[0085]

[0086] S34. Separate the real and imaginary parts of the above formula to obtain:

[0087]

[0088] S35, ω h With Δ ω As an unknown variable, and other physical quantities as coefficients, solving the above equations yields:

[0089]

[0090] In the above formula, the calculation of numerical differentiation can be obtained by dividing the difference between the two sampling values ​​by the sampling period, for example:

[0091]

[0092] in, T s is the sampling period of the data stream.

[0093] S4, according to the cavity pressure signal and dynamic load quality factor Q L , non-thermal quench and some thermal quench events were manually annotated.

[0094] In this embodiment, according to the cavity pressure signal and the dynamic loaded quality factor Q L , manually annotate some thermal quench events, including:

[0095] S41, thermal quench and non-thermal quench events can be measured by dynamic load quality factor Q L Distinguish and manually annotate.

[0096] In this embodiment, Figure 6 The black curve shows the dynamic loaded quality factor Q L From the calculation results, it can be seen that the thermal quench event Figure 6 (a) and Figure 6 (c) will cause dynamic load quality factor Q L The decrease is about one order of magnitude. Figure 6 (d) Generally does not affect the dynamic load quality factor Q L .

[0097] S42. In the event of a thermal quench, the thermal quench mode is marked.

[0098] In this embodiment, Figure 5 As shown in (a) and 6 (a), there is no precursor before the occurrence of ordinary thermal quench. Figure 5 (a) The cavity pressure signal and Figure 6 Dynamic loaded quality factor in (a) Q L There is no obvious abnormality in the curve before the thermal quench. Before the thermal quench, a sharp drop in the cavity pressure amplitude caused by the electrical quench will be observed. Figure 6 As shown in (b), the cavity pressure drop caused by electrical quench is generally in the microsecond order, much smaller than thermal quench (generally in the millisecond order). Soft quench events generally cause dynamic load quality factor Q L The curve shows regular fluctuations, and its fluctuation amplitude generally exceeds the steady-state loaded quality factor Q L According to the above principles, thermal quench events are manually labeled into the above three categories.

[0099] S5. Extract features from thermal quench events.

[0100] In this embodiment, features are extracted from the thermal quench event, including:

[0101] S51. Feature extraction Q id Used to distinguish thermal quench events from non-thermal quench events.

[0102] In this embodiment, the feature Q id The calculation process is as follows:

[0103]

[0104] In the formula, ω H is the half bandwidth of the superconducting cavity under stable operation, ω H and ω h The difference is: ω H represents the half bandwidth of the cavity under stable operation, which is generally a constant (does not change with time), while ω h is the more general cavity half-bandwidth, which may vary with time in some cases (e.g., when the cavity experiences thermal runaway). Therefore, when the cavity is in stable operation, the differential equation can be simplified to:

[0105]

[0106] Optionally, press the signal cavity V c , forward voltage V f According to the amplitude and phase angle, it is decomposed into polar coordinates as follows:

[0107]

[0108] Substituting the above physical quantities into the simplified differential equation, we get:

[0109]

[0110] Optionally, decomposing the above equation into real and imaginary parts yields:

[0111]

[0112] Let Δ θ = θ-φ , construct a new function u = ρ· cos( θ-φ ), then the above differential equation can be simplified to:

[0113]

[0114] In discrete signals, r The differential can be obtained by dividing the difference between the two sampling values ​​by the sampling period, that is:

[0115]

[0116] in, T s is the sampling period of the data stream. Substituting the above equation into the discrete form of the cavity differential equation:

[0117]

[0118] Note that this formula uses r c replace r , in this embodiment r represents the normalized measurement of the cavity pressure amplitude, and r c represents the predicted value of the cavity pressure amplitude, r c It can be deduced from the formula. When the cavity is operating stably and there is no detuning, the normalized cavity pressure amplitude is r [ n ] 1. Can be preset rc Initial value of r c [0]=1, because at this time u [ n ] 1 (constantly equal to), assuming T s ω H =0.1, then the recursive process is as follows:

[0119]

[0120] It can be seen that the predicted value of the cavity pressure amplitude during stable operation is r c The measured value of cavity pressure amplitude r should always remain equal. When thermal quench occurs, r c will no longer be equal to r . Let the deviation signal e =| r c ─ r |, then Figure 7 and Fig. 9 As shown, after thermal quench occurs, e Significantly increase; non-thermal quench faults usually do not cause e The deviation signal can be calculated by e After a fault occurs (i.e. Figure 7 middle t = 0) (i.e. Figure 7 The area of ​​the shaded part in the figure is used to distinguish thermal quench events. When the average height of the shaded part exceeds the set value, for example, 0.3, it can be considered that thermal quench occurs. Q id >0.3, it is considered a thermal quench event.

[0121] feature Q id The calculation formula is as follows:

[0122]

[0123] in, T Q is the integration time.

[0124] S52. Feature extraction E id Used to distinguish electrical quench events.

[0125] In this embodiment, E id The calculation process is as follows: The cavity pressure amplitude caused by electrical quenchr The differential signal Δ of the cavity pressure amplitude is calculated by decreasing rapidly between 1 and 10 microseconds, which is less than the sampling period of the DAQ system (usually between 10 and 100 microseconds). r [ n ] will help to identify electrical quench-induced thermal quench events.

[0126]

[0127] Among them, the discrete sequence signal as well as are respectively the cavity pressure amplitude | V c |At two adjacent sampling points t = nT s and t =( n+ 1) T s The value at .

[0128] Among them, the denominator of the above formula is mainly used to r Normalize to ensure the features E id It is always between 0 and 1. Generally, an electrical quench will produce a large E id , generally E id >0.2 can be identified as electrical quench.

[0129] S53. Feature extraction N soft Used to distinguish soft quench events.

[0130] In this embodiment, N soft The meaning and calculation process are as follows:

[0131] like Figure 6 (c) and Figure 8 As shown, soft quench will cause the loaded quality factor Q L and deviation signal e Abnormal fluctuations occur. The fluctuation amplitude of the deviation signal e is generally around 0.2, and the defining characteristic is N soft Statistical Signals e The number of times it enters between the lower boundary (e.g. 0.1) and the upper boundary (e.g. 0.3), N soft Generally between 2 and 11, other types of thermal failure timeouts occur. N soft Less than or equal to 1, Figure 8 In the example shown,N soft =6.

[0132] Furthermore, three characteristics were statistically analyzed Q id , E id and N soft The distribution of events in conventional thermal quench (Q), electrical quench induced thermal quench (EQ), soft quench (SQ) and non-thermal quench (NQ) (see Fig.10 ), which provides a basis for optimizing the feature extraction algorithm.

[0133] In this embodiment, Fig.10 (c) shows that the soft quench (SQ) N soft Distributed between 2 and 11, but there are also individual events in thermal quench (Q) and electrical quench-induced thermal quench (EQ) events. N soft The value is 2, you can try to adjust Figure 8 The upper and lower boundaries in further optimization N soft The distribution of each event. Some parameters of the feature extraction algorithm of the present invention need to rely on expert experience. For example: Fig.10 As shown in (a), Q id The calculation formula for (this feature is mainly used to distinguish between thermal quench and non-thermal quench) contains a parameter T Q , T Q The value can be 10 milliseconds, 100 milliseconds or 1 second. T Q The value can be based on Fig.10 (a) Select. T Q When set to 10 milliseconds, the error bars of Q, EQ, and SQ (that is, Fig.10 The rectangles of different heights represent Q id The error bars of NQ overlap on the ordinate, which may mean that T Q = 10 ms is not a good choice. When it is increased to 200 ms, the error bars hardly overlap on the vertical axis, which means that 200 ms may be a better choice. Fig.10 (c) As shown in the design N soft When it comes to upper and lower boundaries, this boundary can be set to 0.1~0.3, or 0.2~0.4, and also needs to be based onN soft The error bar distribution of the N soft Value range and three other cases (Q, EQ, SQ) N soft The value ranges do not overlap, that is, their error bars do not overlap on the vertical axis.

[0134] Furthermore, principal component analysis (PCA) is applied to reduce the dimension of features and visualize the samples in two-dimensional plane coordinates. The results are as follows: Fig.11 As shown in the figure, clustering, distribution, and correlation in the data can be clearly observed. This visualization helps experts better understand the data and discover potential relationships, which helps to classify the original data set in more detail. At the same time, PCA analysis also helps to find potential errors in manual classification results. For example, Fig.11 The two samples in the red ellipse in the figure were manually classified as common thermal quench, but according to the PCA graph, they are closer to the soft quench cluster. After careful analysis of the original fault data, these two samples were found to be manually labeled errors. Based on this, the samples with incorrect labels were re-labeled, and the corrected labeling results were used for subsequent model training.

[0135] S6. Train offline machine learning models.

[0136] In this embodiment, training an offline machine learning model includes:

[0137] S61. Data set division.

[0138] In this embodiment, the marked fault data is randomly divided into training sets and test sets according to the ratio of 7:3 for each type of fault data. A stratified sampling method is used to ensure that samples of different fault types remain balanced in the divided data sets. Then all training sets and test sets are integrated as the total training set and test set. The total training set is shuffled to disrupt the order of samples to increase the generalization ability of the model. Among them, the data finally input into the machine learning model for training is feature data, and there is a one-to-one correspondence between fault data and feature data. Fault data refers to the original waveform data, such as cavity pressure waveform, forward voltage waveform, etc. Three features can be extracted from each fault data Q id , E id and N soft , and marked as one of SQ, Q, EQ and NQ. The three features are finally involved in the machine learning model training. Q id , Eid and N soft .

[0139] S62. Model training and testing.

[0140] In this embodiment, a random forest algorithm is used to construct a fault diagnosis model, and a grid search method is used to optimize the hyperparameters of the fault diagnosis model. The final hyperparameters are set as follows: number of decision trees: 150; maximum tree depth: 10; minimum number of leaf node samples: 2. This is taken as an example, but not limited to this, and the fault diagnosis model is trained based on the training set.

[0141] S63. Model evaluation.

[0142] In this embodiment, the confusion matrix and F1-Score are evaluated to obtain a fault diagnosis model with accuracy that meets the requirements. Fig.12 As shown in the figure, the confusion matrix and F1-Score matrix are plotted to find that the fault diagnosis model has an accuracy rate of over 94% in judging thermal quench and non-quench faults on the test set. Among the three subtypes of thermal quench faults, the classification accuracy rates of conventional thermal quench, thermal quench caused by electrical quench, and soft quench are 98.8%, 96.77%, and 100%, respectively. At the same time, the F1-Score of the fault diagnosis model for the classification of the three subtypes of thermal quench is above 0.98.

[0143] S7. Online classification of failure modes based on machine learning.

[0144] In this embodiment, online classification of fault modes based on machine learning includes:

[0145] S71. After receiving the trigger signal, first obtain the original signal recorded by the low-level system P t , P f and P r And calculate the conversion coefficient X and Y , the normalized cavity pressure is obtained V c and forward voltage V f .

[0146] S72, according to the cavity pressure V c and forward voltage V f Signal, calculate the predicted value of cavity pressure amplitude r c ,Compare r c The measured cavity pressure amplitude rDeviation e .

[0147] S73, from the signal r and e Three features are extracted from Q id , E id and N soft , input the features into the trained fault diagnosis model to realize fault classification and determine whether the current fault belongs to the four categories: conventional thermal quench (Q), electrical quench-induced thermal quench (EQ), soft quench (SQ), and non-thermal quench (NQ).

[0148] Embodiment 2: Embodiment 1 above provides an online classification method for thermal quench modes of radio frequency superconducting cavities based on machine learning. Correspondingly, this embodiment provides an online classification device for thermal quench modes of radio frequency superconducting cavities based on machine learning. The device provided in this embodiment can implement the online classification method for thermal quench modes of radio frequency superconducting cavities based on machine learning of Embodiment 1, and the device can be implemented by software, hardware, or a combination of software and hardware. For the convenience of description, this embodiment is described in terms of functions divided into various units and described separately. Of course, the functions of each unit can be implemented in the same or multiple software and / or hardware during implementation. For example, the device may include integrated or separate functional modules or functional units to execute the corresponding steps in each method of Embodiment 1. Since the device of this embodiment is basically similar to the method embodiment, the description process of this embodiment is relatively simple, and the relevant parts can refer to the partial description of Embodiment 1. The embodiment of the online classification device for thermal quench modes of radio frequency superconducting cavities based on machine learning provided by the present invention is only schematic.

[0149] Specifically, the present invention also provides an online classification device for radio frequency superconducting cavity thermal quench mode based on machine learning, comprising:

[0150] The voltage signal module is configured to obtain the cavity pressure of the radio frequency superconducting cavity. V c and forward voltage V f ;

[0151] The signal calculation module is configured to calculate the cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ;

[0152] Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ;

[0153] The features to be extracted Q id , E id and N soft A fault diagnosis model is input to realize fault classification, wherein the faults include non-thermal quench faults and thermal quench faults. The thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench and soft quench.

[0154] Embodiment 3: This embodiment provides an electronic device corresponding to the online classification method of RF superconducting cavity thermal quenching mode based on machine learning provided in Embodiment 1. The electronic device may be an electronic device for a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of Embodiment 1.

[0155] like Fig.13 As shown, the electronic device includes a processor, a memory, a communication interface and a bus, and the processor, the memory and the communication interface are connected through the bus to complete mutual communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Component (EISA) bus, etc. The memory stores a computer program that can be run on the processor. When the processor runs the computer program, it executes the method of Embodiment 1. The implementation principle and technical effect are similar to those of Embodiment 1 and will not be repeated here. Those skilled in the art can understand that Fig.13 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computing device to which the scheme of the present application is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0156] In a preferred embodiment, the logic instructions in the above-mentioned memory can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), optical disk and other media that can store program codes.

[0157] In a preferred embodiment, the processor may be a central processing unit (CPU), a digital signal processor (DSP) or other general-purpose processors of various types, which are not limited herein.

[0158] Embodiment 4: This embodiment provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include computer instructions. When the computer instructions are executed by a computer, the computer executes the method provided in the above-mentioned embodiment 1.

[0159] Embodiment 5: This embodiment provides a computer program product. The computer program product may include a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the method provided in the above-mentioned embodiment 1. Its implementation principle and technical effects are similar to those of embodiment 1 and will not be repeated here.

[0160] In a preferred embodiment, the computer-readable storage medium may be a tangible device that holds and stores instructions used by the instruction execution device, such as but not limited to an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. The computer-readable storage medium stores computer program instructions that cause the computer to execute the method provided in the first embodiment.

[0161] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (apparatus), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0162] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0164] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In the description of this specification, the description of reference terms "a preferred embodiment", "further", "specifically", "in the present embodiment", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of this specification. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples without contradiction.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online classification of thermal quench modes of radio frequency superconducting cavities based on machine learning, characterized in that: The method includes: Obtaining the cavity pressure of a radio frequency superconducting cavity V c and forward voltage V f ; According to cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ; Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ; The features to be extracted Q id , E id and N soft Input the fault diagnosis model to realize fault classification. The faults include non-thermal quench faults and thermal quench faults. Thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench and soft quench. Q id It is used to distinguish thermal quench events from non-thermal quench events. The calculation formula is: In the formula, e =| r c ─ r |, T Q is the integration time; E id Used to distinguish electrical quench events, E id The calculation formula is: In the formula, Δ r [n] is the differential signal of the cavity pressure amplitude, ; N soft Used to distinguish soft quench events, N soft Characteristic statistical deviation signal e The number of times the lower and upper boundaries are crossed, N soft Set between 2 and 11, other types of thermal failure timeouts occur, N soft Less than or equal to 1, where the lower boundary is set to 0.1 and the upper boundary is set to 0.

3.

2. The online classification method of radio frequency superconducting cavity thermal quench mode based on machine learning according to claim 1 is characterized in that: The training of the fault diagnosis model uses the random forest algorithm to build the fault diagnosis model, and the grid search method is used to optimize the model hyperparameters.

3. The online classification method of radio frequency superconducting cavity thermal quench mode based on machine learning according to claim 2 is characterized in that: Before training the fault diagnosis model, the process of manually labeling thermal quench events is also included, including: Based on dynamic loaded quality factor Q L Distinguish thermal quench and non-thermal quench events and manually mark them; When a thermal quench event occurs, the thermal quench mode is manually marked: conventional thermal quench has no precursors before the occurrence of thermal quench, and the dynamic load quality factor Q L There is no obvious abnormality in the curve before the thermal quench; the electrical quench-induced thermal quench event is observed before the thermal quench, and the cavity pressure amplitude drops sharply due to the electrical quench; the soft quench event will cause the dynamic load quality factor Q L The curve shows regular fluctuations.

4. The online classification method of radio frequency superconducting cavity thermal quench mode based on machine learning according to claim 2 is characterized in that: Dynamic loaded quality factor Q L The calculation formula is: ； In the formula, ω 0 is the cavity resonance frequency, ω h is the cavity half bandwidth.

5. An online classification device for thermal quench modes of radio frequency superconducting cavities based on machine learning, characterized in that: include: The voltage signal module is configured to obtain the cavity pressure of the radio frequency superconducting cavity. V c and forward voltage V f ; The signal calculation module is configured to calculate the cavity pressure V c and forward voltage V f , calculate the predicted value of cavity pressure amplitude r c , compare the predicted values r c The measured cavity pressure amplitude r Deviation e ; Based on the measured cavity pressure amplitude value r and deviation e Extract features separately Q id , E id and N soft ; The features to be extracted Q id , E id and N soft Input the fault diagnosis model to realize fault classification. The faults include non-thermal quench faults and thermal quench faults. Thermal quench faults include conventional thermal quench, thermal quench caused by electrical quench and soft quench. Q id It is used to distinguish thermal quench events from non-thermal quench events. The calculation formula is: In the formula, e =| r c ─ r |, T Q is the integration time; E id Used to distinguish electrical quench events, E id The calculation formula is: In the formula, Δ r [n] is the differential signal of the cavity pressure amplitude, ; N soft Used to distinguish soft quench events, N soft Characteristic statistical deviation signal e The number of times the lower and upper boundaries are crossed, N soft Set between 2 and 11, other types of thermal failure timeouts occur, N soft Less than or equal to 1, where the lower boundary is set to 0.1 and the upper boundary is set to 0.

3.

6. An electronic device, characterized in that: include: at least one processor; as well as A memory communicatively connected to the processor; wherein, The memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to perform the method according to any one of claims 1 to 4.

7. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include computer instructions for causing a computer to execute the method according to any one of claims 1-4.

Images