Ultrasonic detection method and system for uniformity of solid-liquid two-phase mixture

Abstract

The invention relates to the technical field of ultrasonic detection, in particular to an ultrasonic detection method and system for uniformity of a solid-liquid two-phase mixture. Ultrasonic transducers are mounted at the bottom of the container, and 12 groups of ultrasonic sensors are uniformly distributed on the outer wall; the transducer transmits an ultrasonic signal, and the sensor receives an echo signal passing through the mixture; and inputting the echo signals into a one-dimensional convolutional neural network for classification, and judging the uniformity of the mixture. The neural network comprises a convolution layer, a pooling layer and a full connection layer, and a large convolution kernel design is adopted to reduce the parameter quantity. And through pre-trained sample learning features, rapid online nondestructive testing is realized, and real-time feedback is provided for the production process. The device overcomes the limitation of a traditional sampling method, improves the detection efficiency and accuracy, has comprehensiveness, intelligence and wide applicability, can remarkably improve the production efficiency and the product quality, and is suitable for mixing uniformity detection in the industries of cement, chemical engineering, food and the like.

Description

Technical Field

[0001] The invention relates to the technical field of ultrasonic detection, in particular to an ultrasonic detection method and system for uniformity of a solid-liquid two-phase mixture. Background Art

[0002] At present, there are two main types of methods for detecting the uniformity of solid-liquid two-phase mixtures. One is offline monitoring, which is to take samples from different parts of the solid-liquid two-phase mixture for testing, and indirectly judge the uniformity through a certain property of the mixture (usually density). This method has poor timeliness, is time-consuming and labor-intensive, and cannot make real-time judgments on the uniformity of the solid-liquid two-phase mixture; the other is online detection, which has been widely used in recent years. This technology adds excitation signals such as infrared, gamma rays, electromagnetic and laser to the solid-liquid two-phase mixture to be tested, and characterizes the characteristics of the solid-liquid two-phase mixture to be tested by capturing the changes caused by such signals passing through the detection area. Among them, common online non-contact detection technologies include optical detection technology, radiation attenuation technology, near-infrared analysis technology (NIR), industrial tomography detection (IPT) and capacitance detection technology. However, in industrial production, most solid-liquid two-phase mixtures eventually present optical opacity, which limits the application of optical detection technology and near-infrared analysis technology. For some solid-liquid two-phase mixtures with conductivity, the capacitance method is no longer applicable. Radiation is generally not used in ordinary industrial environments because it is harmful to the human body and inconvenient to use. Industrial tomography technology is expensive and complex, making it difficult to promote its use in small and medium-sized enterprises.

[0003] Deep learning has achieved great success in computer vision, speech recognition, natural language processing, fault diagnosis and other fields. Among them, the most widely used is the convolutional neural network (CNN). Compared with traditional machine learning models, the biggest advantage of CNN is that it combines feature extraction and classification. The network can automatically extract features from input data. In recent years, some scholars have also applied CNN to online detection technology. Through learning and training on large-scale samples, and relying on high-precision classification capabilities and rapid response capabilities, this technology has been widely promoted.

[0004] Due to the lack of convenient and safe online detection technology for the uniformity of solid-liquid two-phase mixtures, it is impossible to obtain its uniformity information in real time. At present, in order to ensure the uniformity of solid-liquid two-phase mixtures, the industry generally adopts the method of extending the mixing time to mechanically improve the uniformity of the material, which greatly reduces production efficiency, is time-consuming and labor-intensive, and the final mixing effect is not satisfactory. At the same time, with the continuous development of industrial technology, fast-response online detection technology has become an indispensable part of enterprises. Therefore, under the background of existing production processes and future new technology development, new methods for the uniformity detection of solid-liquid two-phase mixtures are urgently needed. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is that due to the lack of convenient and safe online detection technology for the uniformity of solid-liquid two-phase mixtures, it is impossible to obtain its uniformity information in real time. At present, in order to ensure the uniformity of solid-liquid two-phase mixtures, the industry generally adopts the method of extending the mixing time to mechanically improve the uniformity of materials, which greatly reduces production efficiency, is time-consuming and labor-intensive, and the final mixing effect is not satisfactory. At the same time, with the continuous development of industrial technology, fast-response online detection technology has become an indispensable link in enterprises.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides an ultrasonic detection method for uniformity of a solid-liquid two-phase mixture, which includes: a processing system includes a solid-liquid mixture container, an ultrasonic transducer, 12 sets of ultrasonic sensors, a signal processing unit, and an integrated control unit;

[0009] The ultrasonic transducer is installed at the bottom of the container, and 12 groups of sensors are evenly distributed on the outer wall of the container. The ultrasonic transducer transmits ultrasonic signals, and the ultrasonic sensor receives echo signals passing through the solid-liquid mixture;

[0010] After processing, the echo signal is input into a one-dimensional convolutional neural network for classification to judge the uniformity of the solid-liquid mixture. The one-dimensional convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer. A large convolution kernel design is used to reduce the number of parameters.

[0011] The one-dimensional convolutional neural network learns features through pre-trained samples to achieve fast online non-destructive detection of the uniformity of solid-liquid two-phase mixtures and provide real-time feedback for the production process.

[0012] As a preferred solution of the ultrasonic detection method for the uniformity of the solid-liquid two-phase mixture of the present invention, wherein: four groups of ultrasonic sensors are arranged equidistantly in the circumferential direction of the solid-liquid two-phase mixture container to be tested, each group of three probes, so that 12 ultrasonic signals can be received; an ultrasonic transducer is arranged at the bottom of the solid-liquid two-phase mixture container, and the container is connected through the probe;

[0013] Twelve groups of ultrasonic sensors receive the ultrasonic signals emitted by the ultrasonic transducer at the bottom of the solid-liquid two-phase mixture container. After further data processing and integration, electrical signals with different amplitudes are generated, which are finally introduced into the one-dimensional convolutional neural network in the integrated control unit for classification to determine the uniformity of the solid-liquid two-phase mixture.

[0014] As a preferred embodiment of the ultrasonic detection method for the uniformity of the solid-liquid two-phase mixture of the present invention, wherein: the ultrasonic coupling agent is applied to the contact surface between the ultrasonic transducer, the sensor probe and the solid-liquid two-phase mixture container, and the solid-liquid two-phase mixture to be tested is poured into the container;

[0015] The integrated control unit outputs a sinusoidal wave driving signal, which is converted into an analog signal through a D / A converter in the signal processing unit, and then passes through a driving circuit and a transmitting circuit in sequence to amplify its power and make the driving signal reach the ultrasonic transducer probe at the bottom of the solid-liquid two-phase mixture container, and finally converts the signal into mechanical vibration, thereby transmitting an ultrasonic signal into the container;

[0016] The ultrasonic signal generated by the ultrasonic transducer passes through the solid-liquid two-phase mixture to be tested in the container, and is finally received by 12 groups of ultrasonic sensor probes evenly distributed on the outer wall of the container, and the signal is converted into an ultrasonic echo signal;

[0017] The ultrasonic echo signal is amplified by the amplifier circuit in the signal processing unit, filtered by the filter, and then sampled by the A / D converter with a sampling frequency of 16kHz and a sampling time of 4s, and the sampled data is transmitted to the integrated control unit for further processing;

[0018] In the integrated control unit, the 12 returned time domain signals are merged and sorted, and then imported into the trained and tested one-dimensional convolutional neural network for classification to determine the uniformity of the solid-liquid two-phase mixture.

[0019] As a preferred embodiment of the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture of the present invention, the process of training and testing the one-dimensional convolutional neural network samples is as follows:

[0020] Select 10 groups of solid-liquid two-phase mixtures with different uniformities, and pour the first 8 groups into containers for testing;

[0021] Each time, the detected signal is introduced into a one-dimensional convolutional neural network for training. The 12 sets of data are integrated into one piece of data from bottom to top according to their relative positions. In order to increase the number of samples, the 12 sets of data are spliced ​​in a random order, that is, the four sets of data located at different positions in the container can form a set of data chains in different orders. Each set of data has 64,000 data points, and the set of samples is labeled, +1 represents good uniformity, and -1 represents poor uniformity.

[0022] In order to avoid chance, solid-liquid two-phase mixtures with good and poor uniformity are tested and trained alternately;

[0023] After model training, the remaining two groups of solid-liquid two-phase mixtures with different uniformity were poured into containers for testing;

[0024] The signal after each detection is introduced into the trained one-dimensional convolutional neural network for testing. The data integration method and the sample capacity expansion method and steps are consistent each time the signal after detection is introduced into the one-dimensional convolutional neural network for training. Finally, the uniformity evaluation result of the solid-liquid two-phase mixture given by the observation system is considered to be successful if the error is within 5%;

[0025] After a lot of training and learning, the uniformity of the solid-liquid two-phase mixture can be detected in real time if the error is within an acceptable range, and materials with poor mixing uniformity can be reworked.

[0026] As a preferred embodiment of the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture of the present invention, the convolution layer is composed of a feature map obtained by convolution operation of a plurality of convolution kernels. The input of a one-dimensional convolutional neural network is a one-dimensional vector, so the convolution kernel of the network is also one-dimensional. The one-dimensional convolution operation can be expressed as:

[0027]

[0028] in, are the input and bias of the kth neuron in the lth layer, is the convolution kernel between the i-th neuron in the l-1th layer and the k-th neuron in the lth layer, is the output of the i-th neuron in the l-1th layer, N l-1 is the number of neurons in the l-1th layer, conv1D(·) is a one-dimensional convolution operation;

[0029] In order to increase the nonlinearity of the network, an activation function is usually applied after each convolution operation. In order to prevent the gradient from disappearing and accelerate the convergence of the network, the activation function uses a rectified linear unit, and the expression is:

[0030] f(x)=max(0,x)

[0031] Therefore, the final output of each neuron in the convolutional layer is:

[0032]

[0033] Pooling is a downsampling operation that uses the overall features of the adjacent area of ​​a certain position as the output of the network at that position to reduce the size of the feature map and the number of networks. At the same time, it can effectively avoid network overfitting. The pooling adopts maximum value pooling, that is, taking the maximum value in the adjacent area of ​​a certain position as the final output of the position:

[0034]

[0035] Among them, H is the width of the convolution kernel;

[0036] In order to avoid the change of internal covariates and gradient diffusion of neurons in each layer and accelerate the convergence of the network, batch normalization can be used to standardize each hidden layer so that each layer has the same Gaussian distribution with a mean of 0 and a variance of 1:

[0037]

[0038] Among them, μ B , are the mean and standard deviation of the batch data, ε is an infinite decimal to avoid the denominator being 0, x i , x′ i are the input and intermediate output of the i-th neuron before batch normalization, γ and β are the scale transformation factor and offset factor introduced to restore the network expression ability, respectively, which are obtained by network learning, y i is the final output of the neuron after batch normalization.

[0039] As a preferred embodiment of the ultrasonic detection method for the uniformity of the solid-liquid two-phase mixture of the present invention, the process of building the one-dimensional convolutional neural network structure model is as follows:

[0040] In a two-dimensional convolutional neural network, a small convolution kernel is generally used. When multiple layers of small convolution kernels are cascaded, not only can the same receptive field as a large convolution kernel be obtained, but the nonlinear expression ability of the network is also enhanced, while the number of convolution parameters is reduced. When the number of input and output channels is , the cascaded 2-layer 3×3 convolution kernel has the same receptive field as the 5×5 convolution kernel, but the number of parameters of the cascaded 2-layer 3×3 convolution kernel is 18C. 2 , while the number of parameters of a 5×5 convolution kernel is 25C 2 ;

[0041] Considering that the convolution kernel size of a one-dimensional convolutional neural network is a scalar, and the convolution kernel of a two-dimensional convolutional neural network is a two-dimensional matrix, the above rule does not apply to one-dimensional convolutional neural networks. When cascading 2 layers of convolution kernels with a size of 3, the number of parameters is 6C. 2 , and the number of parameters of a convolution kernel of size 5 is 5C 2 , so a large convolution kernel should be used in one-dimensional convolutional neural networks.

[0042] As a preferred solution of the ultrasonic detection method for the uniformity of the solid-liquid two-phase mixture of the present invention, wherein: a large convolution kernel should be used in the one-dimensional convolutional neural network; including the first convolution layer using a large convolution kernel of size 5, and then the convolution kernel size is reduced layer by layer, and the step size of all convolution layers is 1, and the padding pixel padding is selected as the "same" mode, so that the size of the feature map remains unchanged during the propagation process;

[0043] Each convolutional layer is followed by a maximum pooling layer with a kernel size of 2 and a stride of 2 to halve the size of the feature map;

[0044] The number of channels of the first convolution layer is 64, the number of channels of the second convolution layer is 128, and the output size of the last pooling layer is 25×128, which is flattened into a 3200-dimensional vector form, and finally connected to two fully connected layers for classification. In addition, in order to avoid overfitting of the network, a batch normalization layer is added after each convolution operation and before the activation function.

[0045] In a second aspect, an embodiment of the present invention provides a system for sequentially acquiring multi-resource inertia and primary frequency modulation auxiliary services, which includes:

[0046] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture as described in the first aspect of the present invention are implemented.

[0047] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture as described in the first aspect of the present invention are implemented.

[0048] The beneficial effects of the present invention are as follows: This scheme realizes all-round, rapid and accurate detection of the uniformity of solid-liquid two-phase mixtures. The large convolution kernel design reduces the number of network parameters and improves the generalization ability of the model. The pre-trained network realizes real-time online non-destructive testing and provides instant feedback for the production process. The splicing of out-of-order data enhances sample diversity and improves the robustness of the model. The batch normalization layer effectively avoids the problem of overfitting. Fully automated operation reduces human errors and ensures the consistency of results. This method overcomes the limitations of traditional detection, provides reliable support for industrial production quality control, and is expected to significantly improve production efficiency and product quality, bringing huge economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 A flow chart of an ultrasonic testing method for homogeneity of a solid-liquid two-phase mixture;

[0051] Figure 2 A computer device diagram for an ultrasonic testing method for homogeneity of a solid-liquid two-phase mixture;

[0052] Figure 3 A schematic diagram of an ultrasonic testing system for an ultrasonic testing method for uniformity of a solid-liquid two-phase mixture;

[0053] Figure 4 A schematic diagram of the one-dimensional convolutional neural network structure of an ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture;

[0054] Figure 5 One-dimensional convolutional neural network model diagram for ultrasonic detection method of uniformity of solid-liquid two-phase mixture. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0058] Example 1

[0059] Reference Figure 1-2 , which is the first embodiment of the present invention, provides an ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture, comprising:

[0060] S100: The processing system includes a solid-liquid mixture container, an ultrasonic transducer, 12 sets of ultrasonic sensors, a signal processing unit and an integrated control unit;

[0061] Preferably, the ultrasonic transducer is a piezoelectric ceramic transducer with a frequency of 200 kHz. The ultrasonic sensor is composed of a piezoelectric chip.

[0062] Preferably, the signal processing unit comprises an amplifying circuit, a filtering circuit, an A / D and D / A converter, a driving circuit, a transmitting circuit and a receiving circuit.

[0063] Preferably, the integrated control system includes a controller, a central integrated controller and controlled equipment, and the controlled equipment includes an LCD screen display system, a lighting system and audio-visual equipment.

[0064] Preferably, the one-dimensional convolutional neural network is composed of a convolutional layer, a pooling layer, a fully connected layer and an output layer, wherein the convolutional layer and the pooling layer are alternately distributed. Figure 3 shown.

[0065] S200: The ultrasonic transducer is installed at the bottom of the container, and 12 sets of sensors are evenly distributed on the outer wall of the container. The ultrasonic transducer transmits ultrasonic signals, and the ultrasonic sensor receives echo signals passing through the solid-liquid mixture;

[0066] S201: 4 groups of ultrasonic sensors are arranged equidistantly in the circumferential direction of the solid-liquid two-phase mixture container to be tested, each group having three probes, so that 12 ultrasonic signals can be received; an ultrasonic transducer is arranged at the bottom of the solid-liquid two-phase mixture container, and the container is connected through the probe;

[0067] Twelve groups of ultrasonic sensors receive the ultrasonic signals emitted by the ultrasonic transducer at the bottom of the solid-liquid two-phase mixture container. After further data processing and integration, electrical signals with different amplitudes are generated, which are finally introduced into the one-dimensional convolutional neural network in the integrated control unit for classification to determine the uniformity of the solid-liquid two-phase mixture.

[0068] S202: Apply ultrasonic coupling agent to the contact surface between the ultrasonic transducer, the sensor probe and the solid-liquid two-phase mixture container, and pour the solid-liquid two-phase mixture to be tested into the container;

[0069] The integrated control unit outputs a sinusoidal wave driving signal, which is converted into an analog signal through a D / A converter in the signal processing unit, and then passes through a driving circuit and a transmitting circuit in sequence to amplify its power and make the driving signal reach the ultrasonic transducer probe at the bottom of the solid-liquid two-phase mixture container, and finally converts the signal into mechanical vibration, thereby transmitting an ultrasonic signal into the container;

[0070] The ultrasonic signal generated by the ultrasonic transducer passes through the solid-liquid two-phase mixture to be tested in the container, and is finally received by 12 groups of ultrasonic sensor probes evenly distributed on the outer wall of the container, and the signal is converted into an ultrasonic echo signal;

[0071] The ultrasonic echo signal is amplified by the amplifier circuit in the signal processing unit, filtered by the filter, and then sampled by the A / D converter with a sampling frequency of 16kHz and a sampling time of 4s, and the sampled data is transmitted to the integrated control unit for further processing;

[0072] In the integrated control unit, the 12 returned time domain signals are merged and sorted, and then imported into the trained and tested one-dimensional convolutional neural network for classification to determine the uniformity of the solid-liquid two-phase mixture.

[0073] S300: inputting the processed echo signal into a one-dimensional convolutional neural network for classification to judge the uniformity of the solid-liquid mixture. The one-dimensional convolutional neural network includes a convolution layer, a pooling layer, and a fully connected layer, and adopts a large convolution kernel design to reduce the number of parameters;

[0074] S301: The process of training and testing one-dimensional convolutional neural network samples is as follows:

[0075] Select 10 groups of solid-liquid two-phase mixtures with different uniformities, and pour the first 8 groups into containers for testing;

[0076] Each time, the detected signal is introduced into a one-dimensional convolutional neural network for training. The 12 sets of data are integrated into one piece of data from bottom to top according to their relative positions. In order to increase the number of samples, the 12 sets of data are spliced ​​in a random order, that is, the four sets of data located at different positions in the container can form a set of data chains in different orders. Each set of data has 64,000 data points, and the set of samples is labeled, +1 represents good uniformity, and -1 represents poor uniformity.

[0077] In order to avoid chance, solid-liquid two-phase mixtures with good and poor uniformity are tested and trained alternately;

[0078] After model training, the remaining two groups of solid-liquid two-phase mixtures with different uniformity were poured into containers for testing;

[0079] The signal after each detection is introduced into the trained one-dimensional convolutional neural network for testing. The data integration method and the sample capacity expansion method and steps are consistent each time the signal after detection is introduced into the one-dimensional convolutional neural network for training. Finally, the uniformity evaluation result of the solid-liquid two-phase mixture given by the observation system is considered to be successful if the error is within 5%;

[0080] After a lot of training and learning, the uniformity of the solid-liquid two-phase mixture can be detected in real time if the error is within an acceptable range, and materials with poor mixing uniformity can be reworked.

[0081] S302: The convolution layer is composed of feature maps obtained by convolution operations of several convolution kernels. The input of a one-dimensional convolutional neural network is a one-dimensional vector, so the convolution kernel of the network is also one-dimensional. The one-dimensional convolution operation can be expressed as:

[0082]

[0083] in, are the input and bias of the kth neuron in the lth layer, is the convolution kernel between the i-th neuron in the l-1th layer and the k-th neuron in the lth layer, is the output of the i-th neuron in the l-1th layer, N l-1 is the number of neurons in the l-1th layer, conv1D(·) is a one-dimensional convolution operation;

[0084] In order to increase the nonlinearity of the network, an activation function is usually applied after each convolution operation. In order to prevent the gradient from disappearing and accelerate the convergence of the network, the activation function uses a rectified linear unit, and the expression is:

[0085] f(x)=max(0,x)

[0086] Therefore, the final output of each neuron in the convolutional layer is:

[0087]

[0088] Pooling is a downsampling operation. It uses the overall features of the adjacent area of ​​a certain position as the output of the network at that position to reduce the size of the feature map and the number of networks. It can also effectively avoid network overfitting. Pooling uses maximum value pooling, that is, taking the maximum value in the adjacent area of ​​a certain position as the final output of that position:

[0089]

[0090] Among them, H is the width of the convolution kernel;

[0091] In order to avoid the change of internal covariates and gradient diffusion of neurons in each layer and accelerate the convergence of the network, batch normalization can be used to standardize each hidden layer so that each layer has the same Gaussian distribution with a mean of 0 and a variance of 1:

[0092]

[0093] Among them, μ B , are the mean and standard deviation of the batch data, ε is an infinite decimal to avoid the denominator being 0, x i , x′ i are the input and intermediate output of the i-th neuron before batch normalization, γ and β are the scale transformation factor and offset factor introduced to restore the network expression ability, respectively, which are obtained by network learning, y i is the final output of the neuron after batch normalization.

[0094] S303: The process of building a one-dimensional convolutional neural network structure model is as follows:

[0095] In a two-dimensional convolutional neural network, a small convolution kernel is generally used. When multiple layers of small convolution kernels are cascaded, not only can the same receptive field as a large convolution kernel be obtained, but the nonlinear expression ability of the network is also enhanced, while the number of convolution parameters is reduced. When the number of input and output channels is , the cascaded 2-layer 3×3 convolution kernel has the same receptive field as the 5×5 convolution kernel, but the number of parameters of the cascaded 2-layer 3×3 convolution kernel is 18C. 2 , while the number of parameters of a 5×5 convolution kernel is 25C 2 ;

[0096] Considering that the convolution kernel size of a one-dimensional convolutional neural network is a scalar, and the convolution kernel of a two-dimensional convolutional neural network is a two-dimensional matrix, the above rule does not apply to one-dimensional convolutional neural networks. When cascading 2 layers of convolution kernels with a size of 3, the number of parameters is 6C. 2 , and the number of parameters of a convolution kernel of size 5 is 5C 2 , so a large convolution kernel should be used in one-dimensional convolutional neural networks.

[0097] Large convolution kernels should be used in one-dimensional convolutional neural networks; including the first convolution layer using a large convolution kernel of size 5, and then the convolution kernel size is reduced layer by layer, and the step size of all convolution layers is 1, and the padding pixel is selected as the "same" mode, so that the size of the feature map remains unchanged during the propagation process;

[0098] Each convolutional layer is followed by a maximum pooling layer with a kernel size of 2 and a stride of 2 to halve the size of the feature map;

[0099] The number of channels in the first convolution layer is 64, the number of channels in the second convolution layer is 128, and the output size of the last pooling layer is 25×128, which is flattened into a 3200-dimensional vector form. Finally, two fully connected layers are connected for classification. In addition, in order to avoid overfitting of the network, a batch normalization layer is added after each convolution operation and before the activation function. The final overall neural network model is shown in the figure below. Figure 5 shown.

[0100] S400: A one-dimensional convolutional neural network learns features from pre-trained samples to achieve fast online non-destructive testing of the uniformity of solid-liquid two-phase mixtures, providing real-time feedback for the production process.

[0101] Furthermore, the present embodiment also provides a multi-resource inertia and primary frequency modulation auxiliary service sequential acquisition system, including an assembly module, a processing system including a solid-liquid mixture container, an ultrasonic transducer, 12 sets of ultrasonic sensors, a signal processing unit and an integrated control unit;

[0102] Receiving module, the ultrasonic transducer is installed at the bottom of the container, and 12 groups of sensors are evenly distributed on the outer wall of the container. The ultrasonic transducer transmits ultrasonic signals, and the ultrasonic sensor receives echo signals passing through the solid-liquid mixture;

[0103] The classification module processes the echo signal and inputs it into a one-dimensional convolutional neural network for classification to determine the uniformity of the solid-liquid mixture. The one-dimensional convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer, and adopts a large convolution kernel design to reduce the number of parameters.

[0104] In the output module, the one-dimensional convolutional neural network learns features through pre-trained samples to achieve fast online non-destructive detection of the uniformity of solid-liquid two-phase mixtures and provide real-time feedback for the production process.

[0105] This embodiment also provides a computer device, which is suitable for the ultrasonic detection method of the uniformity of a solid-liquid two-phase mixture, and includes a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the ultrasonic detection method of the uniformity of a solid-liquid two-phase mixture proposed in the above embodiment.

[0106] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0107] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture as proposed in the above embodiment is implemented.

[0108] In summary, the present invention achieves all-round detection of the mixture in the entire container by arranging ultrasonic transducers at the bottom of the container and evenly distributing 12 groups of ultrasonic sensors on the outer wall of the container. This arrangement overcomes the limitations of traditional sampling detection, improves the accuracy and representativeness of detection, and effectively avoids misjudgment caused by improper sampling.

[0110] The one-dimensional convolutional neural network is used to process and classify the echo signal, realizing the intelligent judgment of the uniformity of the solid-liquid mixture. Compared with the traditional signal processing method, this method can automatically extract signal features without the need to manually design complex feature extraction algorithms, greatly improving the efficiency and accuracy of detection.

[0111] The use of large convolution kernel design in the one-dimensional convolutional neural network effectively reduces the number of network parameters. This design not only reduces the computational complexity, but also improves the generalization ability of the model, making the model better adapted to different types of solid-liquid mixtures.

[0112] By pre-training the network, fast online non-destructive testing is achieved. This method avoids the time delay of traditional offline testing and can provide real-time feedback for the production process, which is conducive to timely adjustment of production parameters and improvement of production efficiency and product quality.

[0113] The training was performed by splicing 12 sets of data in a random order, which significantly increased the number and diversity of samples. This data enhancement method enables it to adapt to more diverse detection scenarios.

[0114] Adding a batch normalization layer after the convolutional layer effectively avoids the overfitting problem of the network. This design improves the generalization ability of the model, allowing it to maintain good performance when facing new unknown samples.

[0115] The entire testing process does not require human intervention and is fully automated, which not only improves testing efficiency, but also reduces the impact of human errors and ensures the consistency and reliability of test results.

[0116] In summary, this solution achieves fast, accurate and non-destructive detection of the uniformity of solid-liquid two-phase mixtures through innovative sensor arrangement, advanced neural network algorithms and intelligent data processing methods. This method not only overcomes the limitations of traditional detection methods, but also provides real-time and reliable detection results, providing strong support for quality control in industrial production processes, and is expected to significantly improve production efficiency and product quality, bringing significant economic benefits to related industries.

[0117] Example 2

[0118] Reference Figure 2 - Figure 5 , which is the second embodiment of the present invention, and this embodiment provides an ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0119] This example uses cement slurry as the detection object to verify the effectiveness of this solution.

[0120] Test preparation: PO 42.5 ordinary Portland cement with a water-cement ratio of 0.5 was selected. 1000g of cement was mixed with 500ml of water to prepare cement slurry. A cylindrical container with a diameter of 30cm and a height of 40cm was prepared as a test container. A 200kHz piezoelectric ceramic ultrasonic transducer was installed at the center of the bottom of the container. Four groups of ultrasonic sensors were installed equidistantly along the circumferential direction of the outer wall of the container, with 3 sensors in each group, located at 10cm, 20cm and 30cm from the bottom, for a total of 12 sensors.

[0121] Detection process: Pour the prepared cement slurry into a container. Start the detection system, and the ultrasonic transducer transmits an ultrasonic signal with a frequency of 200kHz. 12 sensors simultaneously receive the ultrasonic signal passing through the cement slurry, with a sampling frequency of 16kHz and a sampling time of 4s. The 12 sets of collected signals are input into a pre-trained one-dimensional convolutional neural network for processing.

[0122] Network structure: The input layer is 12×64000 one-dimensional data. The first convolution layer uses 64 5×1 convolution kernels with a step size of 1 and padding of "same". The second convolution layer uses 128 3×1 convolution kernels. Each convolution layer is followed by a maximum pooling layer with a pooling kernel size of 2×1 and a step size of 2. Finally, two fully connected layers are used to output the uniformity classification results.

[0123] In order to compare the effect of this method, the traditional sampling method and the conductivity method were used for testing. The traditional sampling method took 3 samples from 5 locations (four corners and the center) in the container and measured their density. The conductivity method arranged 5 pairs of electrodes in the container to measure the conductivity change.

[0124] The test was carried out at 0min (just after pouring), 5min, 10min, 15min, and 20min of stirring, and each method was tested 3 times to take the average value. The test results are shown in the following table:

[0125] Stirring time (min) The results of this method Sampling method uniformity (%) Conductivity method uniformity (%) 0 Uneven 68.5 71.2 5 Uneven 82.3 85.7 10 Basically uniform 93.6 92.1 15 Uniform 97.8 96.5 20 Uniform 98.2 97.9

[0126] Data Analysis:

[0127] From the table data, it can be seen that the test results of this scheme are well consistent with those of the traditional sampling method and the conductivity method, which proves the reliability of this method. The specific analysis is as follows:

[0128] Detection sensitivity: At the initial stage of stirring (0min and 5min), this method can accurately determine that the mixture is in an "uneven" state, which is consistent with the low uniformity values ​​of the sampling method and the conductivity method. This shows that this method has a high sensitivity to small changes in the uniformity of the mixture.

[0129] Judgment accuracy: After stirring for 10 minutes, this method judged it as "basically uniform", and the uniformity of the corresponding sampling method and conductivity method was between 92% and 94%, which was in a critical state. This shows that this method can accurately capture the transition state in the mixing process, which is conducive to more precise control of the mixing process.

[0130] Stability: After stirring for 15 and 20 minutes, the method was stably judged as "uniform", which is consistent with the high uniformity of more than 96% of the sampling method and the conductivity method. This shows that the method can give a stable and reliable judgment after the mixture reaches a stable state.

[0131] Detection efficiency: This method only takes 4 seconds to complete each test, while the traditional sampling method requires sampling, weighing, calculation and other steps, which often takes 15-20 minutes; although the conductivity method has a fast detection speed, the initial electrode installation is complicated and is easily affected by the change of cement slurry conductivity. In comparison, this method greatly improves the detection efficiency.

[0132] Comprehensiveness: This method uses 12 sensors to perform all-round detection at different locations in the container, overcoming the misjudgment caused by improper sampling positions in the sampling method, and also avoiding the defect of the conductivity method that mainly reflects local uniformity. This ensures the representativeness and reliability of the test results.

[0133] Intelligence: This method uses a one-dimensional convolutional neural network to automatically extract features and classify them without manual intervention. Compared with the traditional method that requires professional judgment, this method significantly reduces the reliance on the professional quality of operators and improves the objectivity and consistency of the test results.

[0134] Adaptability: Although this embodiment takes cement slurry as an example, this method is applicable to various solid-liquid two-phase mixtures in principle. Traditional sampling methods often require different sampling and measurement schemes to be designed according to different materials, and the conductivity law is difficult to apply to non-conductive mixtures. Therefore, this method has a wider range of applicability.

[0135] In summary, this scheme shows obvious advantages in detection sensitivity, judgment accuracy, detection efficiency, comprehensiveness, intelligence and adaptability. These advantages enable this method to better meet the needs of real-time and accurate monitoring of mixing uniformity in industrial production, and are expected to significantly improve production efficiency and product quality. Especially in industries such as cement, chemical industry, and food that have high requirements for mixing uniformity, the application of this method will bring considerable economic benefits.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for ultrasonically detecting the homogeneity of a solid-liquid two-phase mixture, characterized in that: include, The processing system includes a solid-liquid mixture container, an ultrasonic transducer, 12 sets of ultrasonic sensors, a signal processing unit and an integrated control unit; The ultrasonic transducer is installed at the bottom of the container, and 12 sets of sensors are evenly distributed on the outer wall of the container. The ultrasonic transducer transmits ultrasonic signals, and the ultrasonic sensor receives echo signals passing through the solid-liquid mixture; After processing, the echo signal is input into a one-dimensional convolutional neural network for classification to judge the uniformity of the solid-liquid mixture. The one-dimensional convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer. A large convolution kernel design is used to reduce the number of parameters. The one-dimensional convolutional neural network learns features through pre-trained samples to achieve fast online non-destructive detection of the uniformity of solid-liquid two-phase mixtures and provide real-time feedback for the production process.

2. The ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to claim 1, characterized in that: Four groups of ultrasonic sensors are arranged equidistantly in the circumferential direction of the solid-liquid two-phase mixture container to be tested, each group having three probes, so that 12 ultrasonic signals can be received; an ultrasonic transducer is arranged at the bottom of the solid-liquid two-phase mixture container, and the container is connected through the probe; Twelve groups of ultrasonic sensors receive the ultrasonic signals emitted by the ultrasonic transducer at the bottom of the solid-liquid two-phase mixture container. After further data processing and integration, electrical signals with different amplitudes are generated, which are finally introduced into the one-dimensional convolutional neural network in the integrated control unit for classification to determine the uniformity of the solid-liquid two-phase mixture.

3. The ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to claim 2, characterized in that: The ultrasonic coupling agent is applied to the contact surface between the ultrasonic transducer, the sensor probe and the solid-liquid two-phase mixture container, and the solid-liquid two-phase mixture to be tested is poured into the container; The integrated control unit outputs a sinusoidal wave driving signal, which is converted into an analog signal through a D / A converter in the signal processing unit, and then passes through a driving circuit and a transmitting circuit in sequence to amplify its power and make the driving signal reach the ultrasonic transducer probe at the bottom of the solid-liquid two-phase mixture container, and finally converts the signal into mechanical vibration, thereby transmitting an ultrasonic signal into the container; The ultrasonic signal generated by the ultrasonic transducer passes through the solid-liquid two-phase mixture to be tested in the container, and is finally received by 12 groups of ultrasonic sensor probes evenly distributed on the outer wall of the container, and the signal is converted into an ultrasonic echo signal; The ultrasonic echo signal is amplified by the amplifier circuit in the signal processing unit, filtered by the filter, and then sampled by the A / D converter with a sampling frequency of 16kHz and a sampling time of 4s, and the sampled data is transmitted to the integrated control unit for further processing; In the integrated control unit, the 12 returned time domain signals are merged and sorted, and then imported into the trained and tested one-dimensional convolutional neural network for classification to determine the uniformity of the solid-liquid two-phase mixture.

4. The ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to claim 3, characterized in that: The process of training and testing the one-dimensional convolutional neural network sample is as follows: Select 10 groups of solid-liquid two-phase mixtures with different uniformities, and pour the first 8 groups into containers for testing; Each time, the detected signal is introduced into a one-dimensional convolutional neural network for training. The 12 sets of data are integrated into one piece of data from bottom to top according to their relative positions. In order to increase the number of samples, the 12 sets of data are spliced ​​in a random order, that is, the four sets of data located at different positions in the container can form a set of data chains in different orders. Each set of data has 64,000 data points, and the set of samples is labeled, +1 represents good uniformity, and -1 represents poor uniformity. In order to avoid accidental results, solid-liquid two-phase mixtures with good and poor uniformity were tested and trained alternately; After model training, the remaining two groups of solid-liquid two-phase mixtures with different uniformity were poured into containers for testing; The signal after each detection is introduced into the trained one-dimensional convolutional neural network for testing. The data integration method and the sample capacity expansion method and steps are consistent each time the signal after detection is introduced into the one-dimensional convolutional neural network for training. Finally, the uniformity evaluation result of the solid-liquid two-phase mixture given by the observation system is considered to be successful if the error is within 5%; After a lot of training and learning, the uniformity of the solid-liquid two-phase mixture can be detected in real time if the error is within an acceptable range, and materials with poor mixing uniformity can be reworked.

5. The ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to claim 4, characterized in that: The convolution layer is composed of feature maps obtained by convolution operations of several convolution kernels. The input of a one-dimensional convolutional neural network is a one-dimensional vector, so the convolution kernel of the network is also one-dimensional. The one-dimensional convolution operation can be expressed as: in, are the input and bias of the kth neuron in the lth layer, is the convolution kernel between the i-th neuron in the l-1th layer and the k-th neuron in the lth layer, is the output of the i-th neuron in the l-1th layer, N l-1 is the number of neurons in the l-1th layer, conv1D(·) is a one-dimensional convolution operation; In order to increase the nonlinearity of the network, an activation function is usually applied after each convolution operation. In order to prevent the gradient from disappearing and accelerate the convergence of the network, the activation function uses a rectified linear unit, and the expression is: f(x)=max(0,x) Therefore, the final output of each neuron in the convolutional layer is: Pooling is a downsampling operation that uses the overall features of the adjacent area of ​​a certain position as the output of the network at that position to reduce the size of the feature map and the number of networks. At the same time, it can effectively avoid network overfitting. The pooling adopts maximum value pooling, that is, taking the maximum value in the adjacent area of ​​a certain position as the final output of the position: Among them, H is the width of the convolution kernel; In order to avoid the change of internal covariates and gradient diffusion of neurons in each layer and accelerate the convergence of the network, batch normalization can be used to standardize each hidden layer so that each layer has the same Gaussian distribution with a mean of 0 and a variance of 1: Among them, μ B , are the mean and standard deviation of the batch data, ε is an infinite decimal to avoid the denominator being 0, x i , x′ i are the input and intermediate output of the i-th neuron before batch normalization, γ and β are the scale transformation factor and offset factor introduced to restore the network expression ability, respectively, which are obtained by network learning, y i is the final output of the neuron after batch normalization.

6. The ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to claim 5, characterized in that: The one-dimensional convolutional neural network structure model building process is: In a two-dimensional convolutional neural network, a small convolution kernel is generally used. When multiple layers of small convolution kernels are cascaded, not only can the same receptive field as a large convolution kernel be obtained, but the nonlinear expression ability of the network is also enhanced, while the number of convolution parameters is reduced. When the number of input and output channels is , the cascaded 2-layer 3×3 convolution kernel has the same receptive field as the 5×5 convolution kernel, but the number of parameters of the cascaded 2-layer 3×3 convolution kernel is 18C. 2 , while the number of parameters of a 5×5 convolution kernel is 25C 2 ; Considering that the convolution kernel size of a one-dimensional convolutional neural network is a scalar, and the convolution kernel of a two-dimensional convolutional neural network is a two-dimensional matrix, the above rule does not apply to one-dimensional convolutional neural networks. When cascading 2 layers of convolution kernels with a size of 3, the number of parameters is 6C. 2 , and the number of parameters of a convolution kernel of size 5 is 5C 2 , so a large convolution kernel should be used in one-dimensional convolutional neural networks.

7. The ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to claim 6, characterized in that: A large convolution kernel should be used in the one-dimensional convolutional neural network, including the first convolution layer using a large convolution kernel of size 5, and then the convolution kernel size is reduced layer by layer, and the step size of all convolution layers is 1, and the padding pixel is selected as the "same" mode, so that the size of the feature map remains unchanged during the propagation process; Each convolutional layer is followed by a maximum pooling layer with a kernel size of 2 and a stride of 2 to halve the size of the feature map; The number of channels of the first convolution layer is 64, the number of channels of the second convolution layer is 128, and the output size of the last pooling layer is 25×128, which is flattened into a 3200-dimensional vector form, and finally connected to two fully connected layers for classification. In addition, in order to avoid overfitting of the network, a batch normalization layer is added after each convolution operation and before the activation function.

8. An ultrasonic detection system for the uniformity of a solid-liquid two-phase mixture, based on the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to any one of claims 1 to 7, characterized in that: Also includes, Assembling the module, the processing system includes a solid-liquid mixture container, an ultrasonic transducer, 12 sets of ultrasonic sensors, a signal processing unit and an integrated control unit; Receiving module, the ultrasonic transducer is installed at the bottom of the container, and 12 groups of sensors are evenly distributed on the outer wall of the container. The ultrasonic transducer transmits ultrasonic signals, and the ultrasonic sensor receives echo signals passing through the solid-liquid mixture; The classification module processes the echo signal and inputs it into a one-dimensional convolutional neural network for classification to determine the uniformity of the solid-liquid mixture. The one-dimensional convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer, and adopts a large convolution kernel design to reduce the number of parameters. In the output module, the one-dimensional convolutional neural network learns features through pre-trained samples to achieve fast online non-destructive detection of the uniformity of solid-liquid two-phase mixtures and provide real-time feedback for the production process.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the ultrasonic detection method for the uniformity of a solid-liquid two-phase mixture according to any one of claims 1 to 7 are implemented.

Images