A Fault Diagnosis Method for the Lift Screw of a Czochralski Silicon Single Crystal Furnace Based on Multi-Source Signals

Abstract

The present application proposes a method for diagnosing faults in the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals, belonging to the technical field of fault diagnosis of Czochralski silicon single crystal furnaces. It includes: collecting vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace; respectively performing information division and normalization processing on all the collected information to obtain multiple vibration normalization data sets and multiple displacement normalization data sets; constructing fused data information using all the normalization data sets; successively performing coarse-grained reconstruction, time series reconstruction, permutation entropy value calculation, and normalization operations on the fused data information to obtain the fused entropy value corresponding to each scale factor; extracting a fused entropy value feature set from all the fused entropy values, training and validating a support vector machine model to obtain a fault diagnosis model; and using the fault diagnosis model to diagnose faults in the lifting lead screw of the Czochralski silicon single crystal furnace. The present application can improve the accuracy and efficiency of fault diagnosis.

Description

Technical Field

[0001] This application relates to the technical field of fault diagnosis of Czochralski silicon single crystal furnaces, and particularly to a fault diagnosis method for the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals, belonging to the item of classification number G06F18 / 213. Background Art

[0002] At present, silicon single crystals are widely used in the industrial field, and the Czochralski method has become the mainstream method for preparing silicon single crystals. The entire preparation process is carried out in a dedicated Czochralski silicon single crystal furnace. During the preparation process, the crucible lifting mechanism plays an important role. It is responsible for precisely controlling the position of the crucible during the growth of the silicon single crystal to maintain the stability of the liquid level of the silicon solution and the uniformity of the thermal field, thereby ensuring that the prepared silicon single crystal has high quality and high yield. The core transmission component of the crucible lifting mechanism is the lifting lead screw, and the vertical movement of the crucible can be driven by rotating the lifting lead screw. However, long-term and high-load operations may cause the lifting lead screw to malfunction due to problems such as insufficient lubrication, wear, or pitting, and these problems may all have an adverse impact on the quality of the silicon single crystal. Therefore, effective fault diagnosis and maintenance of the lifting lead screw are crucial for ensuring the stable operation of the Czochralski silicon single crystal furnace and preparing high-quality silicon single crystals.

[0003] Traditional fault diagnosis methods for lifting lead screws often rely on the data information collected by a single type of sensor. For example, vibration signals are collected through vibration sensors, and displacement signals are collected using displacement sensors, etc. Therefore, the complex characteristics of faults cannot be comprehensively captured. To improve the accuracy and robustness of fault diagnosis, the vibration signals and displacement signals of the lifting lead screw in the Czochralski silicon single crystal furnace can be fused, and the fusion result can be used as the input data set, and combined with the permutation entropy value algorithm for fault detection. However, such methods have the following problems:

[0004] First, using single-scale entropy values for fault detection often makes it difficult to effectively capture the dynamic changes of signals at different time scales;

[0005] Second, using multi-scale entropy values for fault detection can quantify the complexity of signals at multiple time scales. However, in the traditional coarse-graining process, as the scale factor increases, the length of the coarse-grained sequence will significantly decrease; and at higher time scales, the sequence may become too short to ensure the reliability of the entropy value;

[0006] Third, the calculation of entropy values often depends on the recognition of patterns in the sequence. For relatively short sequences, it is very likely that insufficient information is provided to accurately evaluate these patterns, thus affecting the stability and reliability of the entropy value.

[0007] Therefore, it is necessary to propose a solution to improve one or more problems existing in the above-mentioned related technical solutions.

[0008] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present application, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0009] The embodiments of the present application provide a method for diagnosing faults of the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals. The method includes the following steps:

[0010] Collect the vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace;

[0011] Perform information partitioning and normalization processing on the vibration information and the displacement information in sequence, and respectively obtain a vibration normalization data set and a displacement normalization data set corresponding to each scale factor;

[0012] Use all the vibration normalization data sets and all the displacement normalization data sets to construct fused data information;

[0013] Perform coarse-grained reconstruction, time series reconstruction, permutation entropy value calculation and normalization operations on the fused data information in sequence, and respectively obtain a fused entropy value corresponding to each scale factor;

[0014] Extract multiple fused entropy value features from all the fused entropy values to form a fused entropy value feature set, and use the fused entropy value feature set to train and verify a support vector machine model to obtain a fault diagnosis model;

[0015] Use the fault diagnosis model to diagnose faults of the lifting lead screw of the Czochralski silicon single crystal furnace.

[0016] In an exemplary embodiment of the present application, the step of performing information partitioning and normalization processing on the vibration information and the displacement information in sequence, and respectively obtaining a vibration normalization data set and a displacement normalization data set corresponding to each scale factor includes:

[0017] Divide the vibration information into multi-stage vibration information sequences, and divide the displacement information into multi-stage displacement information sequences, where the multi-stage vibration information sequences include an acceleration stage vibration subsequence, a constant speed stage vibration subsequence, and a deceleration stage vibration subsequence; the multi-stage displacement information sequences include an acceleration stage displacement subsequence, a constant speed stage displacement subsequence, and a deceleration stage displacement subsequence;

[0018] Perform the normalization processing on the multi-stage vibration information sequence and the multi-stage displacement information sequence respectively, and obtain the vibration normalization data set and the displacement normalization data set corresponding to each scale factor respectively.

[0019] In an exemplary embodiment of the present application, the step of constructing the fusion data information by using all the vibration normalization data sets and all the displacement normalization data sets includes:

[0020] Calculate the vibration fault sensitivity coefficient by using all the vibration normalization data sets, and perform vibration weight assignment on all the vibration normalization data sets by using the vibration fault sensitivity coefficient;

[0021] Calculate the displacement fault sensitivity coefficient by using all the displacement normalization data sets, and perform displacement weight assignment on all the displacement normalization data sets by using the displacement fault sensitivity coefficient;

[0022] Fuse all the vibration normalization data sets and all the displacement normalization data sets by using the results of all the vibration weight assignments and all the displacement weight assignments to obtain the fusion data information.

[0023] In an exemplary embodiment of the present application, the expression of the fault sensitivity coefficient is:

[0024] (1)

[0025] Wherein, represents the fault sensitivity coefficient, and the fault sensitivity coefficient is the vibration fault sensitivity coefficient or the displacement fault sensitivity coefficient, represents the normalization data set of the lifting lead screw in the current working state, represents the normalization data set of the lifting lead screw in the normal working state, and the normalization data set is the vibration normalization data set or the displacement normalization data set;

[0026] The expression of the fusion data information is:

[0027] (2)

[0028] Wherein, represents the fusion data information at the th moment, represents the vibration weight of the vibration normalization data set at the th moment, represents the vibration normalization data set at the th moment, represents the displacement weight of the displacement normalization data set at the th moment, Indicates the displacement normalization dataset at the moment.

[0029] In an exemplary embodiment of the present application, the steps of performing coarse-grained reconstruction, time series reconstruction, permutation entropy value calculation, and normalization operation on the fused data information in sequence to obtain the fused entropy value corresponding to each scale factor include:

[0030] Performing the coarse-grained reconstruction on the fused data information by using a sliding coarse-grained method to obtain the coarse-grained sequences corresponding to each scale factor respectively;

[0031] Performing the time series reconstruction on all the coarse-grained sequences respectively to obtain the time series corresponding to each scale factor respectively;

[0032] Arranging all the time series in ascending order respectively according to the order from small to large to obtain the symbol sequences corresponding to each scale factor respectively, calculating the probability of occurrence of each permutation mode of all the symbol sequences, and calculating the permutation entropy value corresponding to each scale factor respectively according to the probabilities of occurrence of all the permutation modes;

[0033] Performing the normalization operation on all the permutation entropy values respectively to obtain the fused entropy value corresponding to each scale factor respectively;

[0034] Wherein, the length of the coarse-grained sequence is equal to the length of the multi-stage vibration information sequence or the multi-stage displacement information sequence.

[0035] In an exemplary embodiment of the present application, the steps of performing the coarse-grained reconstruction on the fused data information by using a sliding coarse-grained method to obtain the coarse-grained sequences corresponding to each scale factor respectively include:

[0036] Introducing a sliding window during the process of performing the coarse-grained reconstruction, making the size of the sliding window equal to the size of the scale factor, and setting the time step of the sliding window to 1, wherein, the size of the sliding window is represented by and the time step of the sliding window is represented by and the size of the scale factor is represented by ; , ;

[0037] For the given fused data information, the sliding window slides from the starting position of the fused data information, slides one time step each time, until the tail end of the sliding window reaches the end of the fused data information;

[0038] Within the coverage range of the sliding window, perform sliding coarse-graining processing on all the fusion data within the sliding window respectively to obtain the coarse-grained sequences corresponding to each scale factor respectively.

[0039] In an exemplary embodiment of the present application, the expression of the coarse-grained sequence is:

[0040] (3)

[0041] Wherein, represents the -th coarse-grained data obtained after the sliding window slides for the -th time at the scale factor of , represents the length of the fusion data information, represents the end index of the current slide of the sliding window, represents the start index of the last slide of the sliding window, represents the -th fusion data of the fusion data information, represents the -th slide of the sliding window, represents the maximum number of slides of the sliding window.

[0042] In an exemplary embodiment of the present application, the expression of the time series is:

[0043] (4)

[0044] Wherein, represents the -th time series at the scale factor of , represents the delay time, represents the embedding dimension of the time series, represents the -th time series at the scale factor of and the number of all time data in the

[0045] In an exemplary embodiment of the present application, the expression of the symbol sequence is:

[0046] (5)

[0047] Wherein, represents the symbol sequence of the -th permutation method after ascending order, represents the -th permutation method, and the number of all symbol data in the symbol sequence, represents the embedding dimension of the symbol sequence, and each All of the said symbol sequences have in common theoretical permutation ways, denotes the factorial of denotes the number of all actual permutation ways, ;

[0048] The expressions for calculating the probability of each of the said permutation ways of all the said symbol sequences are:

[0049] (6)

[0050] wherein, denotes the probability of the th permutation way of the symbol sequence, denotes that the total probability of all actual permutation ways of the symbol sequence arranged in ascending order is 1;

[0051] The expression for calculating the said permutation entropy value is:

[0052] (7)

[0053] wherein, denotes the permutation entropy value;

[0054] The expression for the said fusion entropy value is:

[0055] (8)

[0056] wherein, denotes the fusion entropy value.

[0057] In an exemplary embodiment of the present application, the steps of extracting a plurality of fusion entropy value features from all the said fusion entropy values, forming a fusion entropy value feature set, and using the fusion entropy value feature set to train and verify a support vector machine model to obtain a fault diagnosis model include:

[0058] Using the principal component analysis method to extract a plurality of the said fusion entropy value features from all the said fusion entropy values to form the fusion entropy value feature set;

[0059] Dividing the fusion entropy value feature set into a training set and a test set, and the ratio of the training set to the test set is 7:3 or 8:2 or 6:4;

[0060] Using the training set to train the support vector machine model, and using the test set to verify the trained support vector machine model to obtain the fault diagnosis model.

[0061] Beneficial effects:

[0062] The present application provides a method for diagnosing faults of the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals, which has at least the following beneficial effects:

[0063] (1) By fusing the vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace, the present application can more fully describe the operating state of the lifting lead screw and enrich the state information of the lifting lead screw;

[0064] (2) By adding a sliding window to the traditional coarse-graining process, the present application designs a sliding coarse-graining method, so as to extract the operating state characteristics of the lifting lead screw corresponding to different scale factors and enhance the resistance to noise;

[0065] (3) By calculating the fusion entropy values corresponding to different scale factors and using the principal component analysis method to extract multiple fusion entropy value features from all the fusion entropy values to form a fusion entropy value feature set, the present application reduces the dimension of the fusion entropy value features and improves the accuracy and efficiency of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application and used together with the specification to explain the principles of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0067] Figure 1 A schematic diagram showing the steps of a method for diagnosing faults of the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals in an exemplary embodiment of the present application;

[0068] Figure 2 Showing in an exemplary embodiment of the present application the scale factor A schematic diagram showing the process of the sliding coarse-graining method;

[0069] Figure 3 A schematic diagram showing the coarse-grained reconstruction using the traditional coarse-graining method at different scale factors in the simulation experiment of the present application;

[0070] Figure 4 A schematic diagram showing the coarse-grained reconstruction using the sliding coarse-graining method at different scale factors in the simulation experiment of the present application;

[0071] Figure 5 A schematic diagram showing the diagnostic results of fault diagnosis using the support vector machine model trained and verified with the traditional entropy value feature set in the simulation experiment of the present application;

[0072] Figure 6Schematic diagram showing the diagnostic results of fault diagnosis using the support vector machine model trained and verified with the fusion entropy value feature set of the present application in the simulation experiment of the present application. Detailed implementation manners

[0073] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art. The features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments.

[0074] In addition, the accompanying drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0075] For this purpose, the present example embodiment provides a method for fault diagnosis of the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals, as Figure 1 shown, the method may include the following steps:

[0076] Step S101: Collect the vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace.

[0077] Step S102: Perform information partitioning and normalization processing on the vibration information and displacement information in sequence, and obtain the vibration normalization data set and displacement normalization data set corresponding to each scale factor respectively.

[0078] Step S103: Use all the vibration normalization data sets and all the displacement normalization data sets to construct fusion data information.

[0079] Step S104: Perform coarse-grained reconstruction, time series reconstruction, permutation entropy value calculation and normalization operations on the fusion data information in sequence, and obtain the fusion entropy value corresponding to each scale factor respectively.

[0080] Step S105: Extract multiple fusion entropy value features from all the fusion entropy values to form a fusion entropy value feature set, and use the fusion entropy value feature set to train and verify the support vector machine model to obtain a fault diagnosis model.

[0081] Step S106: Use the fault diagnosis model to perform fault diagnosis on the lifting lead screw of the Czochralski silicon single crystal furnace.

[0082] An embodiment of the present application proposes a method for diagnosing faults in the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals, which has at least the following beneficial effects:

[0083] (1) By fusing the vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace, the present application can more fully describe the operating state of the lifting lead screw and enrich the state information of the lifting lead screw.

[0084] (2) By adding a sliding window to the traditional coarse-graining process, the present application designs a sliding coarse-graining method, which can extract the operating state characteristics of the lifting lead screw corresponding to different scale factors and enhance the resistance to noise.

[0085] (3) By calculating the fusion entropy values corresponding to different scale factors and using the principal component analysis method to extract multiple fusion entropy value features from all fusion entropy values to form a fusion entropy value feature set, the present application reduces the dimension of the fusion entropy value features and improves the accuracy and efficiency of fault diagnosis.

[0086] Next, a method for diagnosing faults in the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals proposed in this exemplary embodiment will be described in more detail.

[0087] In step S101 of this embodiment, the vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace are collected.

[0088] In this embodiment, a vibration sensor and a displacement sensor are installed on the lifting lead screw of the Czochralski silicon single crystal furnace. The vibration information is collected by the vibration sensor, and the displacement information is collected by the displacement sensor.

[0089] In step S102 of this embodiment, the vibration information and displacement information are respectively subjected to information partitioning and normalization processing to obtain a vibration normalization data set and a displacement normalization data set corresponding to each scale factor.

[0090] Step S102 of this embodiment may include the following sub-steps:

[0091] Sub-step S1021: The vibration information is divided into a multi-stage vibration information sequence, and the displacement information is divided into a multi-stage displacement information sequence. Among them, the multi-stage vibration information sequence includes an acceleration stage vibration sub-sequence, a constant speed stage vibration sub-sequence, and a deceleration stage vibration sub-sequence; the multi-stage displacement information sequence includes an acceleration stage displacement sub-sequence, a constant speed stage displacement sub-sequence, and a deceleration stage displacement sub-sequence.

[0092] Sub-step S1022: The multi-stage vibration information sequence and the multi-stage displacement information sequence are respectively subjected to normalization processing to obtain a vibration normalization data set and a displacement normalization data set corresponding to each scale factor.

[0093] In step S103 of this embodiment, all vibration normalization datasets and all displacement normalization datasets are used to construct fusion data information. Step S103 of this embodiment may include the following sub-steps:

[0094] Sub-step S1031: Calculate the vibration fault sensitivity coefficient using all vibration normalization datasets, and use the vibration fault sensitivity coefficient to perform vibration weight allocation on all vibration normalization datasets respectively, and obtain the vibration weight corresponding to each vibration normalization dataset respectively.

[0095] Sub-step S1032: Calculate the displacement fault sensitivity coefficient using all displacement normalization datasets, and use the displacement fault sensitivity coefficient to perform displacement weight allocation on all displacement normalization datasets respectively, and obtain the displacement weight corresponding to each displacement normalization dataset respectively.

[0096] Sub-step S1033: Use the results of all vibration weight allocations and the results of all displacement weight allocations to fuse all vibration normalization datasets and all displacement normalization datasets to obtain fusion data information. The fusion data information includes multiple fusion data.

[0097] Further, the expression of the fault sensitivity coefficient is:

[0098] (1)

[0099] where represents the fault sensitivity coefficient, and the fault sensitivity coefficient is the vibration fault sensitivity coefficient or the displacement fault sensitivity coefficient, represents the normalization dataset of the lifting lead screw in the current working state, represents the normalization dataset of the lifting lead screw in the normal working state, and the normalization dataset is the vibration normalization dataset or the displacement normalization dataset.

[0100] Further, the expression of the fusion data information is:

[0101] (2)

[0102] where represents the fusion data information at the th moment, represents the vibration weight of the vibration normalization dataset at the th moment, represents the vibration normalization dataset at the th moment, represents the displacement weight of the displacement normalization dataset at the th moment, represents the Displacement normalization data set at a moment.

[0103] In step S104 of this embodiment, the fused data information is successively subjected to coarse-grained reconstruction, time series reconstruction, permutation entropy value calculation, and normalization operations to obtain the fused entropy values corresponding to each scale factor. Step S104 of this embodiment may include the following sub-steps:

[0104] Sub-step S1041: Coarse-grained reconstruction of the fused data information is performed using the sliding coarse-graining method to obtain the coarse-grained sequences corresponding to each scale factor.

[0105] Furthermore, the specific operation process of sub-step S1041 is as follows:

[0106] First, as Figure 2 shown, Figure 2 shows the entire sliding coarse-graining process at the scale factor . During the process of coarse-grained reconstruction, a sliding window is introduced, and the size of the sliding window is set to be equal to the size of the scale factor. The time step of the sliding window is set to 1, where the size of the sliding window is represented by , the time step of the sliding window is represented by , and the size of the scale factor is represented by , , .

[0107] Next, for the given fused data information, the sliding window slides from the starting position of the fused data information, sliding one time step each time until the end of the sliding window reaches the end of the fused data information. As Figure 2 shown, for a sliding window, when the sliding window slides for the first time, the end that first contacts the fused data information is the end of the sliding window, and the end that finally contacts the fused data information is the starting end of the sliding window.

[0108] Finally, as Figure 3 and Figure 4 shown, within the coverage range of the sliding window, sliding coarse-graining processing is respectively performed on all the fused data within the sliding window to obtain the coarse-grained sequences corresponding to each scale factor.

[0109] Figure 3 shows the coarse-grained sequence obtained by using the traditional coarse-graining method; Figure 4 shows the coarse-grained sequence obtained by using the sliding coarse-graining method mentioned in this application. It can be seen that: in the traditional coarse-graining process, as the scale factor increases, the length of the obtained coarse-grained sequence will significantly decrease, and at a higher time scale, the coarse-grained sequence may become very short, unable to ensure the reliability of the fused entropy value. However, the method proposed in this application can overcome this problem.

[0110] Further, the expression of the coarse-grained sequence is:

[0111] (3)

[0112] where represents the -th coarse-grained data obtained after the sliding window slides for the -th time at the scale factor of ; all the obtained coarse-grained data form a coarse-grained sequence; represents the length of the fused data information, represents the end index of the current slide of the sliding window. Since the sliding window starts from the -th data and the size of the sliding window is , so the end index of the current slide is ; represents the start index of the last slide of the sliding window. Since the start index of the last slide of the sliding window cannot exceed the length of the fused data information, therefore, the start index of the last slide of the sliding window is , represents the -th fused data of the fused data information, represents the -th slide of the sliding window, represents the maximum number of slides of the sliding window. In each slide process of the sliding window, each additional slide will increase the coarse-grained sequence obtained after coarse-grained reconstruction. Therefore, the number of slides of the sliding window cannot exceed , otherwise the information of adjacent coarse-grained data will be recalculated. Here, adjacent coarse-grained data may be included in multiple slide processes of the sliding window, resulting in the recalculation of some adjacent coarse-grained data.

[0113] Sub-step S1042: Perform time series reconstruction on all the coarse-grained sequences respectively to obtain the time series corresponding to each scale factor.

[0114] Further, the expression of the time series is:

[0115] (4)

[0116] where represents the -th time series at the scale factor of , represents the delay time, represents the embedding dimension of the time series, represents at the scale factor of When the number of all time data in the

[0117] Sub-step S1043: Arrange all time series in ascending order respectively, obtain the symbol sequences corresponding to each scale factor respectively, calculate the probability of occurrence of each permutation of all symbol sequences respectively, and calculate the permutation entropy value corresponding to each scale factor according to the probability of occurrence of all permutations.

[0118] Furthermore, the symbol sequence expression is:

[0119] (5)

[0120] where represents the symbol sequence of the th permutation after ascending order, represents the number of all symbol data in the symbol sequence of the th permutation, represents the embedding dimension of the symbol sequence, and each -dimensional symbol sequence has a total of theoretical permutation ways, represents the factorial of, represents the number of all actual permutation ways, .

[0121] Here, the theoretical permutation ways refer to that for an m-dimensional symbol sequence, theoretically permutation ways will be generated, but excluding repeated permutations and other situations, there may be only actual permutation ways,

[0122] Even further, in formula (5), , there are a total of different permutation ways, but the probability of occurrence when arranged in ascending order is only one. Therefore, the expression for calculating the probability of occurrence of each permutation of all symbol sequences respectively is:

[0123] (6)

[0124] where represents the probability of occurrence of the th permutation of the symbol sequence, the total probability of occurrence of all actual permutations of the symbol sequence arranged in ascending order is 1.

[0125] At this time, the expression for calculating the permutation entropy value can be defined as:

[0126] (7)

[0127] where represents the permutation entropy value.

[0128] Sub-step S1044: Normalize all the permutation entropy values respectively to obtain the fusion entropy values corresponding to each scale factor.

[0129] Furthermore, when the probabilities of each symbol sequence are equal. At this time, the complexity of all symbol sequences is the highest, and the maximum value of the permutation entropy value is . The expression for the fusion entropy value obtained after normalization is:

[0130] (8)

[0131] where represents the fusion entropy value.

[0132] Furthermore, the length of the coarse-grained sequence is equal to the length of the multi-stage vibration information sequence or the multi-stage displacement information sequence.

[0133] In step S105 of this embodiment, multiple fusion entropy value features are extracted from all the fusion entropy values to form a fusion entropy value feature set, and the support vector machine model is trained and verified using the fusion entropy value feature set to obtain a fault diagnosis model. Step S105 of this embodiment may include the following sub-steps:

[0134] Sub-step S1051: Use the principal component analysis (PCA) to extract multiple fusion entropy value features from all the fusion entropy values to form a fusion entropy value feature set.

[0135] Sub-step S1052: Divide the fusion entropy value feature set into a training set and a test set, and the ratio of the training set to the test set can be 7:3 or 8:2 or 6:4.

[0136] Sub-step S1053: Use the training set to train the support vector machines (SVM) model, and use the test set to verify the trained support vector machine model to obtain a fault diagnosis model.

[0137] In step S106 of this embodiment, the fault diagnosis model is used to diagnose the faults of the lifting screw of the Czochralski silicon single crystal furnace.

[0138] To verify the excellent effect of a fault diagnosis method for the lifting screw rod of a Czochralski silicon single crystal furnace based on multi-source signals proposed in this application, the following simulation experiments were carried out.

[0139] Data source of this simulation experiment: the lifting system of the Czochralski silicon single crystal furnace of Xi'an University of Technology. The vibration information and displacement information of the lifting screw rod were collected during the operation of this system.

[0140] This simulation experiment compared the traditional coarse-graining process with the improved sliding coarse-graining process in this application. The specific parameter settings are as follows:

[0141] The scale factors of both the traditional coarse-graining process and the sliding coarse-graining process were set to: ; The embedding dimensions of the preset parameters of the permutation entropy values were both set to: ; The delay times were both set to: ; The vibration weights were both set to: ; The displacement weights were both set to: .

[0142] This simulation experiment aimed to study the fault diagnosis of the lifting screw rod. A total of three fault categories were identified, namely pitting, wear, and insufficient lubrication, as well as a normal working state. Based on the fusion entropy value and relevant experience in the normal working state, thresholds for the fusion entropy values were set for these three fault categories to distinguish different working states. Subsequently, the extracted feature data were labeled with corresponding tags, and the support vector machine model was trained and verified using these feature data. In this simulation experiment, the training set and the test set were divided in a ratio of 6:4 to construct a fault diagnosis model.

[0143] The simulation results are as shown in Figure 5 and Figure 6 .

[0144] Figure 5 shows the fault recognition results obtained by training and verifying the support vector machine model using the traditional multi-scale permutation entropy value as the fusion entropy value feature set. The fault recognition results show that the accuracy rate of using the traditional fault diagnosis method for fault category recognition reaches 95%, which indicates that this method performs well in recognizing fault categories in the normal working state. However, the recognition effect for fault category 1 is not ideal, which may be attributed to the insufficient number of collected data samples or the loss of key information during the calculation of the multi-scale entropy value.

[0145] Figure 6It shows that the method proposed in this application, that is, the fusion entropy value feature set with improved multi-scale permutation entropy value is used to train and verify the support vector machine model. The fault recognition results show that: the accuracy rate of fault category recognition using the method proposed in this application is significantly improved to 98.95%. Especially, significant progress has been made in the recognition of fault category 1 and fault category 3. This indicates that, compared with the traditional method, the method proposed in this application has obvious advantages in improving the accuracy rate of fault category recognition and can effectively detect some difficult-to-recognize fault categories.

[0146] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of this application, "a plurality" means two or more, unless otherwise specifically defined.

[0147] In the description of this specification, the description with reference to the terms "an embodiment", "some embodiments", "example", "specific example" or "some examples", etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may 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.

[0148] As mentioned above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

[0149] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of this application. This application is intended to cover any variations, uses, or adaptations of this application, which follow the general principles of this application and include the known common knowledge or conventional technical means in the technical field not disclosed in this application.

Claims

1. A method for diagnosing the faults of the lifting lead screw of a Czochralski silicon single crystal furnace based on multi-source signals, characterized in that The method includes the following steps: Collect the vibration information and displacement information of the lifting lead screw of the Czochralski silicon single crystal furnace; Perform information partitioning and normalization processing on the vibration information and the displacement information in sequence, and obtain a vibration normalization data set and a displacement normalization data set corresponding to each scale factor respectively, including: Partition the vibration information into a multi-stage vibration information sequence, and partition the displacement information into a multi-stage displacement information sequence, where the multi-stage vibration information sequence includes an acceleration stage vibration subsequence, a constant speed stage vibration subsequence, and a deceleration stage vibration subsequence; the multi-stage displacement information sequence includes an acceleration stage displacement subsequence, a constant speed stage displacement subsequence, and a deceleration stage displacement subsequence; Perform the normalization processing on the multi-stage vibration information sequence and the multi-stage displacement information sequence respectively, and obtain the vibration normalization data set and the displacement normalization data set corresponding to each scale factor respectively; Use all the vibration normalization data sets and all the displacement normalization data sets to construct fusion data information, including: Calculate the vibration fault sensitivity coefficient using all the vibration normalization data sets, and perform vibration weight assignment on all the vibration normalization data sets using the vibration fault sensitivity coefficient respectively; Calculate the displacement fault sensitivity coefficient using all the displacement normalization data sets, and perform displacement weight assignment on all the displacement normalization data sets using the displacement fault sensitivity coefficient respectively; Use the results of all the vibration weight assignments and the results of all the displacement weight assignments to fuse all the vibration normalization data sets and all the displacement normalization data sets to obtain the fusion data information; Perform coarse-grained reconstruction, time series reconstruction, permutation entropy value calculation, and normalization operations on the fusion data information in sequence, and obtain a fusion entropy value corresponding to each scale factor respectively, including: Perform the coarse-grained reconstruction on the fusion data information using the sliding coarse-grained method to obtain a coarse-grained sequence corresponding to each scale factor respectively; Perform the time series reconstruction on all the coarse-grained sequences respectively to obtain a time series corresponding to each scale factor respectively; Arrange all the time series in ascending order in sequence, obtain a symbol sequence corresponding to each scale factor respectively, calculate the probability of occurrence of each permutation method of all the symbol sequences, and calculate the permutation entropy value corresponding to each scale factor according to the probability of occurrence of all the permutation methods; Perform the normalization operation on all the permutation entropy values respectively to obtain a fusion entropy value corresponding to each scale factor respectively; Wherein, the length of the coarse-grained sequence is equal to the length of the multi-stage vibration information sequence or the multi-stage displacement information sequence; Extract multiple fusion entropy value features from all the fusion entropy values to form a fusion entropy value feature set, and use the fusion entropy value feature set to train and verify a support vector machine model to obtain a fault diagnosis model; Use the fault diagnosis model to perform fault diagnosis on the lifting lead screw of the Czochralski silicon single crystal furnace.

2. The method for diagnosing the faults of the lifting lead screw of the Czochralski silicon single crystal furnace based on multi-source signals according to claim 1, wherein The expression of the fault sensitivity coefficient is: (1) Among them, represents a fault sensitivity coefficient, and the fault sensitivity coefficient is the vibration fault sensitivity coefficient or the displacement fault sensitivity coefficient. represents the normalized data set of the lifting lead screw in the current working state. represents the normalized data set of the lifting lead screw in the normal working state, and the normalized data set is the vibration normalized data set or the displacement normalized data set; The expression of the fused data information is as follows: (2) Among them, represents the fused data information at the th moment, represents the vibration weight of the vibration normalization data set at the th moment, represents the vibration normalization data set at the th moment, represents the displacement weight of the displacement normalization data set at the th moment, represents the displacement normalization data set at the th moment.

3. The method for diagnosing the faults of the lifting lead screw of the Czochralski single crystal furnace based on multi-source signals according to claim 1, characterized in that, The steps of performing the coarse-grained reconstruction on the fused data information by using the sliding coarse-graining method to obtain the coarse-grained sequences corresponding to each of the scale factors respectively include: During the process of performing the coarse-grained reconstruction, a sliding window is introduced, and the size of the sliding window is made equal to the size of the scale factor. The time step of the sliding window is set to 1, where the size of the sliding window is represented by , the time step of the sliding window is represented by , and the size of the scale factor is represented by . , ; For the given fused data information, the sliding window starts sliding from the starting position of the fused data information, sliding one time step each time until the end of the sliding window reaches the end of the fused data information; Within the coverage range of the sliding window, perform sliding coarse-graining processing on all the fused data within the sliding window respectively to obtain the coarse-grained sequences corresponding to each of the scale factors respectively.

4. The method for diagnosing the faults of the lifting lead screw of the CZ silicon single crystal furnace based on multi-source signals according to claim 3, wherein The expression of the coarse-grained sequence is as follows: (3) Among them, represents that when the scale factor is , the th coarse-grained data obtained after the sliding window slides for the th time, represents the length of the fused data information, represents the end index of the current slide of the sliding window, represents the start index of the last slide of the sliding window, represents the th fused data in the fused data information, represents the th slide of the sliding window, represents the maximum number of slides of the sliding window.

5. The method for diagnosing the faults of the lifting lead screw of the CZ silicon single crystal furnace based on multi-source signals according to claim 3, wherein The expression of the time series is as follows: (4) Among them, represents the -th time series at the scale factor of , represents the delay time, represents the embedding dimension of the time series, represents, at the scale factor of , the quantity of all time data in the -th time series.

6. The method for diagnosing the failure of the lifting lead screw of the Czochralski silicon single crystal furnace based on multi-source signals according to claim 3, characterized in that, The expression of the symbol sequence is as follows: (5) Among them, represents the symbol sequence of the th permutation after ascending order, represents the number of all symbol data in the symbol sequence of the th permutation, represents the embedding dimension of the symbol sequence, and each dimension of the symbol sequence has a total of theoretical permutations, represents the factorial of represents the number of all actual permutations, ; The expression for calculating the probability of occurrence of each permutation of all the symbol sequences respectively is as follows: (6) Among them, represents the probability of occurrence of the th permutation of the symbol sequence, indicating that the total probability of occurrence of all actual permutations of the symbol sequence arranged in ascending order is 1; The expression for calculating the permutation entropy value is as follows: (7) Among them, represents the permutation entropy value; The expression of the fusion entropy value is as follows: (8) Among them, represents the fusion entropy value.

7. The method for diagnosing the faults of the lifting lead screw of the Czochralski silicon single crystal furnace based on multi-source signals according to claim 1, wherein, Extract multiple fusion entropy value features from all the fusion entropy values, The steps of forming a fusion entropy value feature set and using the fusion entropy value feature set to train and verify a support vector machine model to obtain a fault diagnosis model include: Use the principal component analysis method to extract multiple fusion entropy value features from all the fusion entropy values to form the fusion entropy value feature set; Divide the fusion entropy value feature set into a training set and a test set, and the ratio of the training set to the test set is 7:3 or 8:2 or 6:4; Use the training set to train the support vector machine model, and use the test set to verify the trained support vector machine model to obtain the fault diagnosis model.

Images