A method, apparatus, device, and storage medium for identifying bad data.

Abstract

This invention discloses a method, apparatus, device, and storage medium for identifying defective data. This method overcomes the shortcomings of traditional logical judgment and statistical methods that use fixed threshold analysis to identify defective data related to the process parameters and quality parameters of energetic propellant grains in warheads. These shortcomings include poor stability and inaccurate identification when continuous, constant defective data lies between the troughs and peaks of normal data oscillations. This reduces the probability of missed detections and improves the accuracy of defective data identification. Compared to traditional deep neural networks (DNNs), long short-term memory neural networks (LSTMs) have advantages in processing time-series data, generally exhibiting higher model accuracy. Furthermore, gating techniques prevent the vanishing or exploding gradient problems that occur with very long input sequences.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, device, and storage medium for identifying bad data. Background Technology

[0002] The warhead is the final damage unit for various munitions and missiles, mainly composed of a casing, explosive charge, detonation device, and safety device. It is one of the most fundamental products of modern defense science and technology industry. The warhead is the embodiment of the destructive effects of weapons and equipment, such as killing and destroying targets. Precision loading of the propellant grain is the foundation and guarantee for achieving efficient damage from the warhead, and is also one of the key aspects of warhead development. As a key damage component of the warhead, the pressed propellant grain directly determines the warhead's precise and efficient damage performance and overload safety. The control parameters of the pressing process directly affect the quality of the propellant grain. Currently, my country's existing warhead propellant grain pressing process control uses either a positioning method or a constant pressure method, lacking research on the coupling relationship between process parameters and molding quality in the pressed propellant grain process. It fails to fully utilize manufacturing process data and cannot form effective control methods to ensure the consistency of key quality parameters such as the size, density, and strength of the propellant grain, thus affecting the warhead's damage effect and operational safety.

[0003] Modern warfare places higher demands on the lethality and destructive efficiency of warheads. Currently, countries worldwide are actively researching and developing new warhead technologies employing various damage mechanisms to enhance the high-efficiency destructive capabilities of munitions and missiles. With the continuous development of precision warhead loading technology, higher requirements are being placed on loading quality and reliability. Traditional constant-pressure loading methods rely excessively on expert experience, making it difficult to establish accurate mechanistic models based on process parameters and molding quality using traditional methods. Traditional technologies can no longer meet the needs of intelligent military equipment production and the development of highly reliable, high-performance warheads. Currently, my country's precision loading technology is transitioning to a forward-looking R&D system. Digital twins, deep learning, and other digital and intelligent manufacturing technologies will gradually replace the existing relatively extensive production model, moving towards an integrated production and research manufacturing platform. In the future, process parameters and quality parameters from the manufacturing process will become a powerful support for the iteration of existing precision loading technologies.

[0004] During the acquisition of process parameters, the collected data invariably suffers from varying degrees of quality issues due to various reasons. The presence of poor data can significantly reduce the accuracy of the mechanistic model, while an accurate mechanistic model is fundamental to the development of warheads towards high reliability and high performance. Identifying and eliminating the few poor data points in the acquired data is a necessary means to improve the mechanistic model. Poor data in process parameters mainly includes issues such as data loss and excessive data deviation. Currently, there are many methods to solve these problems, such as system-based state estimation and data-driven detection methods.

[0005] A data-driven R&D system naturally relies on a large amount of data, which comes from the collection and testing of existing process and quality parameters. Due to the complexity of the data acquisition environment and the influence of many uncertain factors, the collected data all have varying degrees of data quality problems, thereby reducing the accuracy of data-driven mechanism models. In particular, traditional machine learning and deep learning cannot accurately identify continuous, constant, and poor-quality data that exists between the troughs and peaks of normal data oscillations.

[0006] Therefore, how to provide a method for identifying defective data in the compression molding process of warhead energetic propellant grains, so as to reduce the false negative rate of the identification model and improve the accuracy of defective data identification, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of the above problems, the present invention provides a method, apparatus, device, and storage medium for identifying defective data to overcome or at least partially solve the above problems. Addressing the issue of defective data in the warhead energetic propellant grain compression molding process during the transformation of the precision charge R&D system, when using information such as process parameters and quality parameters, a defective data identification method for warhead energetic propellant grain compression molding is established based on the characteristics of defective data in manufacturing process parameters and quality parameters. This provides effective data support for the subsequent intelligent manufacturing process and digital transformation and upgrading of precision warhead charges, thereby meeting development needs.

[0008] This invention provides the following solution:

[0009] A method for identifying bad data includes:

[0010] Acquire the molding process parameters and molding quality parameters of the equipment during the molding of energetic propellant columns;

[0011] The molding process parameters and the molding quality parameters are subjected to feature standardization processing to obtain raw parameter data, which includes standardized molding process parameters and standardized molding quality parameters.

[0012] The standardized molding process parameters and the standardized molding quality parameters are respectively subjected to time-leading misalignment subtraction and time-lag misalignment subtraction to obtain the logic transformation characteristics of leading misalignment subtraction and lag misalignment subtraction.

[0013] The original parameter data, the leading misalignment subtraction logic transformation features, and the lagging misalignment subtraction logic transformation features are input into the bad data identification model based on LSTM neural network so that the bad data identification model can identify bad data or normal data.

[0014] Preferably, the molding process parameters include the pressing pressure value, the rate of increase of the pressing pressure, the pressing speed, the displacement, the temperature, the vacuum degree, the pressure compensation, the number of pressing cycles, the holding time, the flow rate of the hydraulic oil in the cylinder, and the pre-pressurization and depressurization re-pressurization.

[0015] Preferably, the standardized function is as follows:

[0016]

[0017] In the formula: x c i.t For the first i The first process parameter or quality parameter t The measured value at time, when c =1 indicates a process parameter. c =2 indicates a mass parameter; x c i.s For the first i Ideal values ​​for each process parameter or quality parameter; x c i.nol For the first i The standard value of a process parameter or quality parameter.

[0018] Preferably, the timing lead-lag subtraction and timing lag subtraction include:

[0019] Determine the time-series lead sequence data of the molding process parameters or the molding quality parameters and the time-series lag sequence data of the molding process parameters or the molding quality parameters;

[0020] The time-series lead-off subtraction includes the time-series lag sequence data of the molding process parameters or the molding quality parameters minus the time-series lead sequence data of the molding process parameters or the molding quality parameters.

[0021] The time lag subtraction includes the time lead sequence data of the molding process parameters or the molding quality parameters minus the time lag sequence data of the molding process parameters or the molding quality parameters.

[0022] Preferably, the expression for the timing lead sequence data of the molding process parameters or the molding quality parameters is as follows:

[0023]

[0024] The expression for the time-lag sequence data of the molding process parameters or the molding quality parameters is as follows:

[0025]

[0026] In the formula: when c =1 indicates a process parameter. c =2 indicates a mass parameter; x c i.(1,N-1) It is a sequence of process parameters or quality parameters, with the first data point as the starting point and the (N-1)th data point as the ending point, and the sequence length is N-1.

[0027] Preferably: after subtracting the timing lead and timing lag shifts, a logical transformation is performed, and the transformation form is shown in the following equation:

[0028]

[0029] In the formula: L ( x i.t-1 , x i.t This means that when subtracting sequential data with either a lead or a lag, if two adjacent numbers are the same, the result after the logical transformation is 0; if two adjacent numbers are different, the result after the logical transformation is 1.

[0030] Preferably, the bad data identification model based on LSTM neural network optimizes the parameters using a target loss function, which is shown in the following formula:

[0031]

[0032] In the formula: f i For deep neural networks to the first i The feature vector extracted from the data belongs to the _th data _th ... y i kind; W yi T Indicates category as y i The weight vector; k and n These represent the number of categories and the number of samples, respectively. s and m These are the scaling factor and the distance factor, respectively.

[0033] A device for identifying bad data, comprising:

[0034] The parameter acquisition unit is used to acquire the molding process parameters and molding quality parameters of the equipment during the molding of energetic propellant columns;

[0035] A standardization processing unit is used to perform characteristic standardization processing on the molding process parameters and the molding quality parameters to obtain raw parameter data, wherein the raw parameter data includes standardized molding process parameters and standardized molding quality parameters.

[0036] The feature transformation unit is used to perform time-series lead-off subtraction and time-series lag-off subtraction on the standardized molding process parameters and the standardized molding quality parameters, respectively, so as to obtain the lead-off subtraction logic transformation feature and the lag-off subtraction logic transformation feature.

[0037] The data recognition result output unit inputs the original parameter data, the leading misalignment subtraction logic transformation feature, and the lagging misalignment subtraction logic transformation feature into the bad data recognition model based on LSTM neural network, so that the bad data recognition model can identify bad data or normal data.

[0038] A device for identifying defective data, the device comprising a processor and a memory:

[0039] The memory is used to store program code and transmit the program code to the processor;

[0040] The processor is used to execute the steps of the above-described bad data identification method according to the instructions in the program code.

[0041] A computer-readable storage medium for storing program code for performing the steps of the above-described method for identifying bad data.

[0042] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0043] This application provides a method, apparatus, device, and storage medium for identifying defective data. It overcomes the shortcomings of traditional logical judgment and statistical methods that use fixed threshold analysis to identify defective data related to the process parameters and quality parameters of energetic propellant grains in warheads, which suffer from low accuracy and inaccurate identification when continuous, constant defective data lies between the troughs and peaks of normal data oscillations. This reduces the probability of missed detections and improves the accuracy of defective data identification. Compared to traditional deep neural networks (DNNs), long short-term memory neural networks (LSTMs) have advantages in processing time-series data, generally exhibiting higher model accuracy. Furthermore, gating technology avoids the gradient vanishing or gradient exploding problems that occur when the input sequence is very long.

[0044] In addition, the preferred implementation overcomes the shortcomings of existing feature variables constructed based on human experience in terms of representation ability, the inability of traditional machine learning to solve highly complex nonlinear problems, poor generalization ability, and easy false detection. At the same time, compared with the softmax classifier loss function, the improved loss function AM-softmax enables the trained model to generate a larger decision gap when making decisions, ensuring high similarity of features of the same type and large differences of features of different types, thereby further improving the accuracy of data recognition.

[0045] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0047] Figure 1 This is a flowchart of a method for identifying problematic data provided in an embodiment of the present invention;

[0048] Figure 2 This is an architecture diagram of the bad data identification model provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the LSTM network structure provided in an embodiment of the present invention;

[0050] Figure 4 This is a flowchart of the bad data identification technology provided in the embodiments of the present invention;

[0051] Figure 5 This is a schematic diagram of the structure of a faulty data identification device provided in an embodiment of the present invention;

[0052] Figure 6 This is a schematic diagram of the structure of a faulty data identification device provided in an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0054] See Figure 1 This invention provides a method for identifying problematic data, such as... Figure 1 As shown, the method may include:

[0055] S101: Obtain the molding process parameters and molding quality parameters during the molding process of energetic propellant columns by the equipment; specifically, the molding process parameters include the propellant pressure value, the rate of increase of the propellant pressure, the propellant speed, displacement, temperature, vacuum degree, pressure compensation, number of propellant presses, holding time, flow rate of hydraulic oil in the cylinder, and pre-pressurization and depressurization re-pressurization.

[0056] S102: The molding process parameters and the molding quality parameters are subjected to feature standardization processing to obtain raw parameter data, the raw parameter data including standardized molding process parameters and standardized molding quality parameters;

[0057] S103: Perform time-series lead-shift subtraction and time-series lag-shift subtraction on the standardized molding process parameters and the standardized molding quality parameters respectively to obtain the logic transformation characteristics of lead-shift subtraction and lag-shift subtraction.

[0058] S104: Input the original parameter data, the leading misalignment subtraction logic transformation feature, and the lagging misalignment subtraction logic transformation feature into the bad data identification model based on LSTM neural network, so that the bad data identification model can correctly identify bad data or normal data.

[0059] This application provides a method for identifying defective process parameters and quality parameters based on deep learning. This method utilizes a Long Short-Term Memory (LSTM) neural network to establish an identification model for process parameters, quality parameters, and data classification. This model differs from traditional statistical and conditional logic-based identification methods, overcoming the problem that fixed thresholds cannot accurately identify defective data when process parameters change complexly in dynamic processes. Currently, the main types of defective data include zero values, abnormal jumps, and continuous constant values. The root causes of these types of defective data are primarily communication blockages, sensor malfunctions, and interference.

[0060] The specific technical steps for establishing the data recognition model using this method are as follows:

[0061] (1) Using data acquisition technology, process parameters and quality parameters are collected and detected online during the pressing and molding process of energetic materials in the warhead;

[0062] (2) Collect a certain amount of process parameters and quality parameters data and classify them manually. At the same time, process parameters and quality parameters are subtracted by leading and lagging time-series shifts to form new features.

[0063] (3) The process parameters and quality parameters of the warhead energetic charge compression molding process are used as inputs, and the manual classification labels are used as outputs. The deep neural network LSTM is used for coupled learning to form a fast identification model for defective process parameters and quality parameters.

[0064] The warhead energetic propellant grain compression molding process parameters and quality parameter defect data identification framework proposed in this application are as follows: Figure 2 As shown, the specific ideas for implementing the solution are illustrated.

[0065] Regarding the process parameters for the compression molding of energetic materials, important parameters such as compression pressure, pressurization rate, and holding time directly affect the quality of the compressed propellant. To improve the quality of the compressed energetic propellant grains in the warhead and ensure the safety of the compression process, it is necessary to detect and collect data on the following process parameters: compression pressure, pressure rise rate, compression speed, displacement, temperature, vacuum level, pressure compensation, number of compressions, holding time, hydraulic oil flow rate in the cylinder, and pre-compression and depressurization / re-compression. This information or parameter needs to be obtained in real-time through reliable, stable, visualized, and non-destructive testing methods to sense the equipment status and achieve data acquisition for various process parameters.

[0066] The data acquisition of quality parameters is based on detection methods including sampling physicochemical analysis, ultrasound, laser appearance inspection, density analysis, etc., to explore the influence of process parameters in the compression molding process on characteristics such as the density field and crack distribution of the propellant, thereby detecting quality parameters and the occurrence of abnormal phenomena in the molding process in real time.

[0067] The process parameters and quality parameters mentioned in this invention need to be standardized before model training. The standardization function used is shown in equation (1):

[0068] (1)

[0069] In the formula: x c i.t For the first i The first process parameter or quality parameter t The measured value at time, when c =1 indicates a process parameter. c =2 indicates a mass parameter; x c i.s For the first i Ideal values ​​for each process parameter or quality parameter; x c i.nol For the first i The standard value of a process parameter or quality parameter.

[0070] Current data identification methods based on machine learning and deep learning struggle to accurately identify defective data when it falls within a continuous, constant value between the troughs and peaks of normal data oscillations. To address this issue, this invention performs lead-shift subtraction and lag-shift subtraction operations on the time-series data of process parameters and quality parameters, as detailed below:

[0071] (2)

[0072] (3)

[0073] Equation (2) represents the time-series lead sequence data of process parameters or quality parameters, and Equation (3) represents the time-series lag sequence data of process parameters or quality parameters. c =1 indicates a process parameter. c =2 indicates a mass parameter; x c i.(1,N-1) It is a sequence of process parameters or quality parameters, with the first data point as the starting point and the (N-1)th data point as the ending point, and the sequence length is N-1.

[0074] The timing lead-lag subtraction operation refers to: Equation (3) minus Equation (2), and the timing lag subtraction operation refers to: Equation (2) minus Equation (3). After performing lead-lag and lag subtraction on the original process and quality parameter sequence data, logical transformation is still required. L ( x c i.t-1 , x c i.t The specific transformation is as follows:

[0075] (4)

[0076] Equation (4) indicates that when the sequence data is subtracted by leading or lagging shift, if two adjacent numbers are the same, the result after the logical transformation is 0; if two adjacent numbers are different, the result after the logical transformation is 1.

[0077] Besides the collected process parameters and quality parameters sequences serving as input features to the LSTM network, the result of sequence lead-shift subtraction and logical transformation will be used as a new input feature, and the result of sequence lag-shift subtraction and logical transformation will be used as another new feature for training. Therefore, the input features of the LSTM network model used for defective data identification are: original parameter data, lead-shift subtraction logical transformation features, and lag-shift subtraction logical transformation features, which can be represented as:

[0078] (5)

[0079] In the formula: I Represents the entire input feature matrix, when c =1 indicates a process parameter. c =2 indicates quality parameters; the first row of the matrix represents the original process and quality parameter characteristics collected; the second row represents the characteristics of the original parameters after leading-shift subtraction and logical transformation; the third row represents the characteristics of the original parameters after lagging-shift subtraction and logical transformation. Therefore, each column of the matrix represents three different features of the sequence at a certain time point.

[0080] As can be seen from the input features, the defective data identification model provided in this application requires not only the data itself but also information from the two adjacent data sets before and after it to identify whether a data set is defective. To achieve rapid identification of defective data, this invention uses an LSTM network to perform deep learning on three different time-series features of process parameters and quality parameters. The LSTM network structure is as follows: Figure 3 As shown. LSTM neural networks are a variant of recurrent neural networks (RNNs). Traditional BP networks use fully connected layers, with no connections between neurons in each layer, thus failing to learn the relationship between the current output of a sequence and previous information. In contrast, LSTM's unique recurrent structure gives it a powerful learning ability for time series. At the same time, LSTM also overcomes the gradient vanishing or exploding problems that occur during the training of traditional RNN networks.

[0081] 3 fusion features I =[ I 1, I 2,..., I t ,..., I N As the input to the LSTM network, then t The input at time t is calculated as follows:

[0082] (6)

[0083] In the formula: W i , W f , W o and W c Input gates i t Forgotten Gate f t Output gate o t and input status Weight vector on; b i ,b f , b o , b c These are the corresponding bias vectors; This represents element-wise multiplication of matrices; σ Here, sigmoid is the function, and tanh is the hyperbolic tangent function; h t The hidden layer features of the LSTM network are mapped to a probability distribution through a softmax classifier to simulate the probability of data classification decisions.

[0084] Because the training objective of the softmax classifier loss function is to maximize the output data of the fully connected layer to ensure that different classes can be separated, significant ambiguity appears at the decision boundary, leading to misclassification of the identified data. To address this issue, this application employs an improved loss function, AM-softmax, for parameter optimization. This loss function generates a larger decision margin and reduces the intra-class margin, thereby improving the classification accuracy. The improved loss function is as follows:

[0085] (7)

[0086] In the formula: f i For deep neural networks to the first i The feature vector extracted from the data belongs to the _th data _th ... y i kind; W yi T Indicates category as y i The weight vector; k and n These represent the number of categories and the number of samples, respectively. s and m These are the scaling factor and the distance factor, respectively.

[0087] Figure 4 The flowchart illustrating the method for identifying data on defective formation of energetic warhead pellets is shown below:

[0088] S1: Data collection and acquisition of energetic propellant column molding process parameters and molding quality parameters based on sensing and detection equipment such as sensors.

[0089] S2: Standardize the molding process parameters and quality parameters according to formula (1).

[0090] S3: The molding process parameters and quality parameters are subtracted by timing advance and timing lag according to equations (2) and (3), and then logically transformed according to equation (4) to form three different feature sequences as shown in equation (5).

[0091] S4: Replace the loss function of the softmax classifier of the LSTM neural network with Equation (7), and take three different feature sequences as input and artificial classification labels as output. Divide the data samples into training set, test set and validation set for training to form a bad data identification model based on LSTM.

[0092] S5: When using the model, the sequence data to be identified is used as input to the trained LSTM model in the data format used during training, and its predicted value is the result of data recognition.

[0093] The method provided in this application uses time-series data and the logical transformation of the difference between two adjacent data points as feature inputs for learning. This solves the problem that traditional machine learning and deep learning cannot accurately identify continuous and constant bad data between the troughs and peaks of normal data oscillations, thereby reducing the false negative rate of the identification model and improving the accuracy of bad data identification.

[0094] Meanwhile, by using the improved loss function AM-softmax to directly take distance as the model optimization objective, the problem of the traditional softmax cross-entropy loss function causing significant ambiguity near the decision boundary and easily leading to the misclassification of some data is avoided.

[0095] The defective data identification method provided in this application overcomes the shortcomings of traditional logical judgment and statistical methods that use fixed threshold analysis to identify defective data of process parameters and quality parameters of warhead energetic propellant columns with low accuracy and cannot accurately identify defective continuous constant data when it is between the trough and peak of normal data oscillation. It reduces the probability of model missed detection and improves the accuracy of defective data identification.

[0096] At the same time, it overcomes the shortcomings of existing feature variables constructed based on human experience in terms of representation ability, and the problems that traditional machine learning cannot solve highly complex nonlinearity, poor generalization ability, and easy false detection. Compared with the softmax classifier loss function, the improved loss function AM-softmax enables the trained model to generate a larger decision gap when making decisions, ensuring high similarity of features of the same type and large differences of features of different types, thereby further improving the accuracy of data recognition.

[0097] In addition, compared with traditional deep neural networks (DNNs), long short-term memory neural networks have more advantages in processing time series data. The model accuracy is generally higher, and the gating technique avoids the gradient vanishing or gradient explosion problems that occur when the input sequence is very long.

[0098] See Figure 5 This application embodiment can also provide a bad data identification device, such as... Figure 5 As shown, the device may include:

[0099] The parameter acquisition unit 501 is used to acquire the molding process parameters and molding quality parameters of the equipment during the molding process of energetic propellant columns;

[0100] The standardization processing unit 502 is used to perform characteristic standardization processing on the molding process parameters and the molding quality parameters to obtain raw parameter data, the raw parameter data including standardized molding process parameters and standardized molding quality parameters;

[0101] The feature transformation unit 503 is used to perform time-series lead-off subtraction and time-series lag-off subtraction on the standardized molding process parameters and the standardized molding quality parameters, respectively, so as to obtain the lead-off subtraction logic transformation feature and the lag-off subtraction logic transformation feature.

[0102] The data recognition result output unit 504 inputs the original parameter data, the leading misalignment subtraction logic transformation feature, and the lagging misalignment subtraction logic transformation feature into the bad data recognition model based on LSTM neural network, so that the bad data recognition model can identify bad data or normal data.

[0103] This application embodiment can also provide a bad data identification device, characterized in that the device includes a processor and a memory:

[0104] The memory is used to store program code and transmit the program code to the processor;

[0105] The processor is used to execute the steps of the above-described bad data identification method according to the instructions in the program code.

[0106] like Figure 6 As shown, the defective data identification device provided in this application embodiment may include: a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, memory 11, and communication interface 12 all communicate with each other through the communication bus 13.

[0107] In this embodiment, the processor 10 may be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0108] The processor 10 can call programs stored in the memory 11. Specifically, the processor 10 can execute operations in the embodiments of the bad data identification method.

[0109] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions:

[0110] Acquire the molding process parameters and molding quality parameters of the equipment during the molding of energetic propellant columns;

[0111] The molding process parameters and the molding quality parameters are subjected to feature standardization processing to obtain raw parameter data, which includes standardized molding process parameters and standardized molding quality parameters.

[0112] The standardized molding process parameters and the standardized molding quality parameters are respectively subjected to time-leading misalignment subtraction and time-lag misalignment subtraction to obtain the logic transformation characteristics of leading misalignment subtraction and lag misalignment subtraction.

[0113] The original parameter data, the leading misalignment subtraction logic transformation features, and the lagging misalignment subtraction logic transformation features are input into the bad data identification model based on LSTM neural network so that the bad data identification model can identify bad data or normal data.

[0114] In one possible implementation, the memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function (such as file creation or data read / write). The data storage area may store data created during use, such as initialization data.

[0115] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0116] Communication interface 12 can be an interface for the communication module, used to connect with other devices or systems.

[0117] Of course, it should be noted that, Figure 6The structure shown does not constitute a limitation on the faulty data identification device in the embodiments of this application. In practical applications, the faulty data identification device may include more than Figure 6 More or fewer components as shown, or combinations of certain components.

[0118] This application embodiment may also provide a computer-readable storage medium for storing program code for performing the steps of the above-described method for identifying bad data.

[0119] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0120] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0121] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0122] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for identifying problematic data, characterized in that, include: Acquire the molding process parameters and molding quality parameters of the equipment during the molding of energetic propellant columns; The molding process parameters and the molding quality parameters are subjected to feature standardization processing to obtain raw parameter data, which includes standardized molding process parameters and standardized molding quality parameters. The standardized molding process parameters and the standardized molding quality parameters are respectively subjected to time-leading misalignment subtraction and time-lag misalignment subtraction to obtain the logic transformation characteristics of leading misalignment subtraction and lag misalignment subtraction. The original parameter data, the leading misalignment subtraction logic transformation features, and the lagging misalignment subtraction logic transformation features are input into the bad data identification model based on LSTM neural network so that the bad data identification model can identify bad data or normal data.

2. The method for identifying problematic data according to claim 1, characterized in that, The molding process parameters include the pressing pressure value, the rate of increase of the pressing pressure, the pressing speed, the displacement, the temperature, the vacuum degree, the pressure compensation, the number of pressing cycles, the holding time, the flow rate of the hydraulic oil in the cylinder, and the pre-pressurization and depressurization re-pressurization.

3. The method for identifying problematic data according to claim 1, characterized in that, The standardized function is shown in the following equation: In the formula: x c i.t For the first i The first process parameter or quality parameter t The measured value at time, when c =1 indicates a process parameter. c =2 indicates a mass parameter; x c i.s For the first i Ideal values ​​for each process parameter or quality parameter; x c i.nol For the first i The standard value of a process parameter or quality parameter.

4. The method for identifying problematic data according to claim 1, characterized in that, The timing lead-lag subtraction and timing lag subtraction include: Determine the time-series lead sequence data of the molding process parameters or the molding quality parameters and the time-series lag sequence data of the molding process parameters or the molding quality parameters; The time-series lead-off subtraction includes the time-series lag sequence data of the molding process parameters or the molding quality parameters minus the time-series lead sequence data of the molding process parameters or the molding quality parameters. The time lag subtraction includes the time lead sequence data of the molding process parameters or the molding quality parameters minus the time lag sequence data of the molding process parameters or the molding quality parameters.

5. The method for identifying problematic data according to claim 4, characterized in that, The expression for the time-lead sequence data of the molding process parameters or the molding quality parameters is as follows: The expression for the time-lag sequence data of the molding process parameters or the molding quality parameters is as follows: In the formula: when c =1 indicates a process parameter. c =2 indicates a mass parameter; x c i.(1,N-1) It is a sequence of process parameters or quality parameters, with the first data point as the starting point and the (N-1)th data point as the ending point, and the sequence length is N-1.

6. The method for identifying problematic data according to claim 5, characterized in that, The timing lead-lag subtraction and timing lag subtraction are followed by a logical transformation, the transformation form of which is shown in the following equation: In the formula: L ( x c i.t-1 , x c i.t This means that when subtracting sequential data with either a lead or a lag, if two adjacent numbers are the same, the result after the logical transformation is 0; if two adjacent numbers are different, the result after the logical transformation is 1.

7. The method for identifying problematic data according to claim 1, characterized in that, The bad data identification model based on LSTM neural network optimizes the parameters using a target loss function, which is shown in the following formula: In the formula: f i For deep neural networks to the first i The feature vector extracted from the data belongs to the _th data _th ... y i kind; W yi T Indicates category as y i The weight vector; k and n These represent the number of categories and the number of samples, respectively. s and m These are the scaling factor and the distance factor, respectively.

8. A device for identifying defective data, characterized in that, include: The parameter acquisition unit is used to acquire the molding process parameters and molding quality parameters of the equipment during the molding of energetic propellant columns; A standardization processing unit is used to perform characteristic standardization processing on the molding process parameters and the molding quality parameters to obtain raw parameter data, wherein the raw parameter data includes standardized molding process parameters and standardized molding quality parameters. The feature transformation unit is used to perform time-series lead-off subtraction and time-series lag-off subtraction on the standardized molding process parameters and the standardized molding quality parameters, respectively, so as to obtain the lead-off subtraction logic transformation feature and the lag-off subtraction logic transformation feature. The data recognition result output unit inputs the original parameter data, the leading misalignment subtraction logic transformation feature, and the lagging misalignment subtraction logic transformation feature into the bad data recognition model based on LSTM neural network, so that the bad data recognition model can identify bad data or normal data.

9. A device for identifying defective data, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the steps of the bad data identification method according to any one of claims 1-7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for performing the steps of the bad data identification method according to any one of claims 1-7.

Images