Ball milling analyzer and control method thereof
By using energy spectrum analysis and recurrent neural networks to process audio signals in the ball mill analyzer, the ball mill capacity is predicted in real time, and the problem of insufficient dynamic feedback in the existing technology is solved, and the system intelligence and automation, adaptability and response speed are improved.
Patent Information
- Application Number
- CN202410968880.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-19
AI Technical Summary
Existing ball mill analyzers cannot dynamically feedback the quality of ore, resulting in poor consistency of the system version, slow response speed, low efficiency, difficulty in quickly adapting to the source or quality changes of ore, and low degree of intelligence.
Design a ball mill analyzer to receive the audio signal of the ball mill device through a radio, use a filter module and an intelligent analysis module for signal processing, combine energy spectrum analysis and cyclic neural network to predict ball mill capacity in real time, and based on the hardness and humidity characteristics of the ore Take control.
Accurate prediction of ball mill capacity is achieved, production efficiency and system response speed is improved, the intelligence and automation of the system is enhanced, the complexity and cost of manual intervention is reduced, and the adaptability and robustness to different ores is improved.
Smart Images

Figure CN118925919A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of ball mill analyzers, and in particular, relates to a ball mill analyzer and a control method thereof. Background Art
[0002] For various types of grinding equipment and ball mills for ore, it is difficult to effectively monitor the internal feed volume due to their working principles. They completely rely on the operator's experience and make adjustments based on the tone and volume released by the grinding equipment. In the specific industrial production process, it is difficult to effectively improve the output quality and operation production efficiency. The existing solution is to use a grinding sound analyzer to imitate the operator's ears, monitor the volume of the grinding sound, and use the DCS system to control the feed to improve the level of operation automation.
[0003] Existing grinding sound analyzers cannot dynamically feedback the quality of materials. For example, the grinding process of ores with different hardness and humidity requires different time; at the collection level, the same input amount will have different audio frequencies and amplitudes; at the application level, the same audio feedback will require different grinding times, and different control strategies need to be adopted for ores of different qualities.
[0004] This adjustment method results in poor version consistency of the system. An effective ball mill control optimization strategy requires designers to have a high level of knowledge in geology, control, and other aspects. For the same production system, when the source or quality of the ore changes, it is difficult to quickly switch parameters and strategies, resulting in slow response, low efficiency, and inability to accurately predict the ball mill capacity, resulting in a low level of intelligence.
[0005] In view of this, the present invention is proposed. Summary of the invention
[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a ball mill analyzer and a control method thereof, which can improve production efficiency and system response speed by predicting and feeding back the ball mill capacity in real time and dynamically according to the grinding sound signal during the ore ball milling process, while taking into account the different hardness and humidity characteristics of the ore.
[0007] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is: the present invention provides a grinding sound analyzer, comprising:
[0008] A microphone is used to receive the audio signal frequency band during the operation of the ball mill;
[0009] A filtering module is used to filter the acquired audio signal frequency band and select the required audio signal frequency band;
[0010] An amplification module, used for amplifying the filtered audio signal frequency band;
[0011] An intelligent analysis module is used to obtain the amplified audio signal frequency band, process and analyze the audio signal frequency band and output prediction;
[0012] An output module is used to convert the output prediction result into an actual control signal and output it;
[0013] The processing and analysis module includes: an energy spectrum calculation module, which performs spectrum analysis on the frequency band of the audio signal, extracts the energy of each frequency band as a feature, and constructs a feature vector in combination with the historical state;
[0014] The model reasoning module uses the constructed feature vector to process and predict the current ball mill capacity.
[0015] Further, the energy spectrum calculation module includes: a spectrum analysis unit, which is used to receive the frequency band of the audio signal and divide the audio signal into a number of fixed frequency band intervals;
[0016] An energy spectrum calculation unit, used for calculating the spectrum integral value within each frequency band interval;
[0017] The feature extraction unit is used to take the spectrum integral value of the energy spectrum calculation module as a feature vector, and combine it with the intermediate state quantity of the previous cycle, use linear transformation and activation function to process the feature vector, and construct the feature vector in combination with the historical state.
[0018] Further, the model reasoning module includes: a recurrent neural network unit, which is used to calculate the state and intermediate state quantity of the current cycle by combining the current input and historical input states using the recurrent neural network unit's recurrent structure; the model reasoning module includes a multi-level recurrent neural network unit to form a fully linked layer structure;
[0019] The prediction output unit takes the output of the recurrent neural network unit as the input of the fully connected layer structure, and obtains the final prediction value through linear transformation and activation function.
[0020] The present invention also discloses a control method applied to the ball mill analyzer, the control method comprising the following steps:
[0021] The receiver receives the audio signal from the ball milling equipment, and filters the audio signal to remove noise and unnecessary interference;
[0022] The intelligent analysis module processes the acquired audio signal frequency bands, extracts the energy of each frequency band as features, and extracts features of the grinding progress of ores with different hardness / humidity based on historical status.
[0023] The model reasoning module uses a recurrent neural network to extract features from energy spectrum data and calculate prediction values;
[0024] The output module converts the calculated predicted value into an actual control signal and outputs the calculation result as a standard signal.
[0025] Furthermore, when the intelligent analysis module processes the acquired audio signal frequency band, it also includes:
[0026] Convert the acquired audio signal into a digital signal;
[0027] The converted digital signal is subjected to fast Fourier transform to convert the digital signal from the time domain to the frequency domain.
[0028] Furthermore, when extracting the energy of each frequency band as a feature and combining the historical state to construct the feature extraction of the grinding progress of ores with different hardness / humidity, it also includes:
[0029] Divide the frequency domain signal into several fixed frequency band intervals and calculate the energy spectrum in each frequency band;
[0030] The calculated energy spectrum is used as subsequent input, and the model reasoning module uses a recurrent neural network to process the discrete energy spectrum data, extract the grinding progress characteristics, and predict the ball mill capacity.
[0031] Furthermore, the model reasoning module uses a recurrent neural network to process discrete energy spectrum data, extract grinding progress features, and predict ball mill capacity, specifically including:
[0032] First, Xt is input into the recurrent neural network as the input for subsequent calculations. The recurrent neural network introduces the historical input state Ct-1, processes the feature vector using the predefined weight matrix and activation function, and calculates the state Ht and intermediate state Ct of the current cycle.
[0033] Further,
[0034] The activation function is Sigmoid; the intermediate state Ct=Sig(W1﹒Xt)+(W2﹒Ct-1);
[0035] The Ht=(Sig(W0﹒Xt)☉(Sig(Sig(W1﹒Xt)+(W2﹒Ct-1)));
[0036] Among them: Ht is the state of the current cycle; Sig is the activation function; W0, W1, W2 are predefined weight matrices; Xt is the input quantity; Ct-1 is the historical input state.
[0037] Furthermore, the recurrent neural network is composed of multiple recurrent neural networks connected in series.
[0038] Furthermore, the output module converts the calculated predicted value into an actual control signal and outputs the calculation result as a standard signal, specifically including:
[0039] The output module outputs Ht according to the state of the current cycle of the recurrent neural network, calculates the final output value Out through the full connection layer, the value is between [0,1], and converts it into a 4~20mA current signal for final output;
[0040] Out=Sig(Sum(Wk));
[0041] Among them: Sig is the activation function, Wk is the final weighted vector before the output of the fully connected layer.
[0042] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:
[0043] (1) The grinding sound analyzer of the present invention can quickly respond to the changes in the operating status of the ball mill by receiving the audio signal of the ball mill in real time, processing and analyzing the audio signal, and accurately predict the ball mill capacity. By accurately predicting the ball mill capacity, the operating parameters of the ball mill, such as feed speed, grinding time, etc., can be optimized, thereby improving production efficiency and product quality. The application of the intelligent analysis module and the model reasoning module makes the management of the ball mill more intelligent and automated, reducing the complexity and cost of manual intervention.
[0044] (2) By dividing the detection frequency band into multiple discrete segments and calculating the energy in each segment, the system can capture the spectral characteristics of the input signal more carefully. This refinement helps to identify the subtle differences in the ore during the grinding process and improve the system's resolution. At the same time, the frequency band processing can reduce the interference of noise in the entire frequency band on the useful signal, and can more effectively extract the characteristics related to the ore grinding state. It can also more flexibly adapt to different types of ores, thereby improving the robustness and versatility of the system.
[0045] (3) The present invention uses a recurrent neural network to process time series data, which can capture the time dependence and dynamic changes in the data. In ore grinding monitoring, the introduction of historical calculations as input can help the RNN network better understand the dynamic changes of the grinding process, thereby improving the accuracy of the prediction.
[0046] (4) By connecting multiple recurrent neural networks in series, the system can build a deeper network structure, thereby having stronger feature extraction and pattern recognition capabilities, which helps to mine deep information in the data and improve the prediction accuracy of the system. Using the fully linked network as the output layer can flexibly map the features extracted by the recurrent neural network to the required output range. The Sigmoid function is used as the activation function, which can limit the output value to between (0,1), facilitating scale conversion and standardization processing, and improving the applicability to different scenarios.
[0047] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] The accompanying drawings are part of the present invention and are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but do not constitute an improper limitation of the present invention. Obviously, the drawings described below are only some embodiments. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the accompanying drawings:
[0049] Figure 1 This is a structural block diagram of a grinding sound analyzer of the present invention;
[0050] Figure 2 This is an overall structural diagram of a grinding sound analyzer of the present invention;
[0051] Figure 3 It is a bar graph of energy spectra of multiple fixed frequency bands in the present invention;
[0052] Figure 4 It is a design diagram of a single recurrent neural network structure in the present invention;
[0053] Figure 5 It is a design diagram of the multi-level recurrent neural network structure in the present invention;
[0054] Figure 6 The present invention is a flow chart of a grinding sound analyzer control method.
[0055] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0056] To make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be described below with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present invention. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, the description of known structures and technologies is omitted to avoid unnecessary confusion of the concepts of the present invention.
[0057] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.
[0058] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.
[0059] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).
[0060] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some boxes or combinations thereof in the block diagrams and / or flow charts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions can create a device for implementing the functions / operations described in these block diagrams and / or flow charts when executed by the processor. The technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). In addition, the technology of the present disclosure can take the form of a computer program product on a computer-readable storage medium storing instructions, which can be used by an instruction execution system or used in conjunction with an instruction execution system.
[0061] The grinding sound analyzer is a typical audio acquisition and analysis system. It collects the sound emitted by the grinding equipment on site, filters and extracts the effective frequency components, and establishes a mapping relationship between the audio intensity and the amount of material stored in the grinding equipment and the particle size of the material stored.
[0062] like Figure 1 to Figure 2 As shown, a grinding sound analyzer described in the present invention includes: a radio receiver, a filtering module, an amplifying module, an intelligent analysis module and an output module.
[0063] The microphone is used to receive the audio signal frequency band during the operation of the ball milling equipment.
[0064] The filtering module is used to filter the acquired audio signal frequency band to select the required audio signal frequency band.
[0065] The amplification module is used to amplify the screened audio signal frequency band.
[0066] The intelligent analysis module is used to obtain the amplified audio signal frequency band, process and analyze the audio signal frequency band and output prediction.
[0067] The output module is used to convert the output prediction result into an actual control signal and output it.
[0068] The processing and analysis module includes an energy spectrum calculation module, which performs spectrum analysis on the frequency bands of the audio signal, extracts the energy of each frequency band as a feature, and constructs a feature vector in combination with the historical state.
[0069] The model reasoning module uses the constructed feature vector to process and predict the current ball mill capacity.
[0070] The present invention firstly installs a microphone near the ball milling equipment to receive the audio signal during the operation of the equipment; then the filtering module performs filtering processing on the received audio signal to remove noise and unnecessary frequency bands, and screen out the audio signal frequency bands closely related to the operation status of the ball milling equipment; the amplification module performs amplification processing on the screened audio signal to improve the signal-to-noise ratio and clarity of the signal, and provide a high-quality signal source for subsequent analysis and prediction; then the energy spectrum calculation module in the intelligent analysis module performs spectrum analysis on the amplified audio signal, extracts the energy of each frequency band as a feature, combines the extracted feature with the historical state data, and constructs a feature vector that can fully reflect the current state of the ball milling equipment, and then uses the constructed feature vector to perform data processing and prediction through the model reasoning module to predict the current ball milling capacity; finally, the output module converts the prediction result obtained by model reasoning into an actual control signal, such as a control signal for adjusting parameters such as feeding speed and grinding time, and these control signals act on the ball milling equipment through corresponding actuators to achieve real-time adjustment and optimization of the operation status of the equipment.
[0071] Through the above-mentioned configuration, the present invention can quickly respond to changes in the operating state of the ball mill equipment and realize accurate prediction of the ball mill capacity. By accurately predicting the ball mill capacity, the operating parameters of the ball mill equipment, such as feed speed, grinding time, etc., can be optimized, thereby improving production efficiency and product quality. The application of the intelligent analysis module and the model reasoning module makes the management of the ball mill equipment more intelligent and automated, reducing the complexity and cost of manual intervention.
[0072] Furthermore, the energy spectrum calculation module includes a spectrum analysis unit, an energy spectrum calculation unit and a feature extraction unit.
[0073] The spectrum analysis unit is used to receive the frequency band of the audio signal and divide the audio signal into a number of fixed frequency band intervals.
[0074] The energy spectrum calculation unit is used to calculate the spectrum integral value within each frequency band interval.
[0075] The feature extraction unit is used to use the spectrum integral value of the energy spectrum calculation module as a feature vector, and in combination with the intermediate state quantity of the previous cycle, use linear transformation and activation function to process the feature vector, and construct the feature vector in combination with the historical state.
[0076] Through the above-mentioned setting, the spectrum analysis unit divides the audio signal into several fixed frequency bands, so that the calculation of the energy spectrum is more precise and detailed, and the operating status of the ball mill is reflected more accurately; the energy spectrum calculation unit realizes the quantitative analysis of the energy of the audio signal by calculating the spectrum integral value in each frequency band interval; the feature extraction unit combines the intermediate state quantity and historical state information of the previous cycle, and further processes the feature vector through linear transformation and activation function, so as to improve the expression ability and prediction accuracy of the feature vector.
[0077] Specifically, after receiving the audio signal frequency band, the spectrum analysis unit uses Fourier transform (or other suitable spectrum analysis methods) to convert the audio signal from the time domain to the frequency domain, and divides the frequency domain signal into several fixed frequency band intervals, which can be set according to the characteristics of the ball milling equipment and the prediction requirements. Then the energy spectrum calculation unit integrates the spectrum signal in each frequency band interval, obtains the spectrum integral value (i.e., energy value) in the frequency band interval, and records the energy value of each frequency band interval to form preliminary energy spectrum data. Finally, the feature extraction unit: takes the spectrum integral value obtained by the energy spectrum calculation unit as part of the input feature, and obtains the intermediate state quantity of the previous cycle at the same time, combines the spectrum integral value and the intermediate state quantity of the previous cycle to form a preliminary feature vector, and then applies linear transformation (such as weight matrix multiplication) to process the preliminary feature vector, uses activation functions (such as ReLU, Sigmoid, etc.) to perform nonlinear transformation on the linearly transformed feature vector, combines historical state data (such as historical ball milling capacity, equipment operation time, etc.), and further processes the feature vector after nonlinear transformation to construct the feature vector finally used for model prediction.
[0078] Furthermore, the model reasoning module includes a recurrent neural network unit and a prediction output unit.
[0079] The recurrent neural network unit is used to utilize the cyclic structure of the recurrent neural network unit, combine the current input and historical input states, and calculate the state and intermediate state quantities of the current cycle; the model reasoning module includes multiple levels of recurrent neural network units to form a fully linked layer structure.
[0080] The prediction output unit uses the output of the recurrent neural network unit as the input of the full-link layer structure, and obtains the final prediction value through linear transformation and activation function.
[0081] Recurrent Neural Network Unit (RNN), through its built-in loop structure, the RNN unit can process data with time series characteristics. At each time step, the RNN unit receives the current input features (feature vectors from the energy spectrum calculation module) and the output state of the previous time step, and calculates the output state of the current time step and the new internal state accordingly.
[0082] The RNN units in the model reasoning module of this application are organized into a multi-level structure. Each level receives the output of the previous level as input and generates its own output. It can capture more complex temporal dependencies and improve the prediction ability of the model. Multiple RNNs are connected in series to form a fully connected layer structure for information transmission.
[0083] The prediction output unit receives the final output of the RNN unit as input and performs a linear transformation on the input (i.e., multiplying it by a weight matrix and adding a bias term) to adjust the contribution of different features to the prediction results.
[0084] Apply an activation function (such as Sigmoid, Softmax, etc., selected according to the specific requirements of the prediction task) to perform a nonlinear transformation on the result after linear transformation to obtain the final prediction value.
[0085] Specifically, Figure 3 As shown, first, the input signal is divided into several fixed frequency bands through spectral analysis, and the energy spectrum of the input signal in the frequency band interval is calculated. The entire frequency band range is 5KHz, the spectrum integral calculation value within 0-100Hz is x0, the spectrum integral within 100-200Hz is x1..., and the spectrum integral within 4.9KHz-5KHz is xn; x1-xn are used as input quantities for subsequent calculations, and the recurrent neural network (RNN) structure is used to predict the current ball mill capacity. By introducing the historical input state Ct-1, the grinding progress of ores with different hardness / humidity is feature extracted.
[0086] Where: Xt=[x0t,x1t…xnt], Ct-1 is the historical input state;
[0087] Ht=(Sig(W0﹒Xt)☉(Sig(Sig(W1﹒Xt)+(W2﹒Ct-1)))
[0088] Where Sig is the activation function Sigmoid;
[0089] The above intelligent analysis module uses a field programmable gate array (FPGA) as a processor to provide parallel computing for the recurrent neural network.
[0090] Figure 6 As shown, the present invention also discloses a control method applied to the ball mill analyzer, and the control method comprises the following steps:
[0091] S1. The microphone receives the audio signal in the ball milling device, and performs filtering on the audio signal to remove noise and unnecessary interference.
[0092] S2. The intelligent analysis module processes the acquired audio signal frequency bands, extracts the energy of each frequency band as features, and extracts features of the grinding progress of ores with different hardness / humidity based on the historical status.
[0093] S3, the model reasoning module uses a recurrent neural network to extract features from energy spectrum data and calculate prediction values.
[0094] S4. The output module converts the calculated predicted value into an actual control signal and outputs the calculation result as a standard signal.
[0095] This application can reflect the status of ore grinding in real time by directly analyzing the audio signal in the ball mill, including changes in key parameters such as hardness and humidity, thereby improving the control accuracy of the grinding process. Combined with the intelligent analysis module and the model reasoning module, it can automatically learn and adapt to changes in different ore characteristics, realize intelligent prediction and control of grinding progress, improve production efficiency and product quality, accurately control the grinding process, avoid energy waste and equipment wear caused by excessive or insufficient grinding, and reduce production costs.
[0096] Specifically, firstly, the audio signal is collected and preprocessed, and the collected audio signal is filtered to remove unnecessary interference such as environmental noise and mechanical vibration to ensure the signal quality. Then, the intelligent analysis module is used to perform spectrum analysis on the processed audio signal, divide the signal into multiple frequency bands, and extract the energy of each frequency band as the characteristic value. The characteristic value can reflect the changes in the physical properties of the ore during the grinding process, such as hardness, humidity, etc. Then, combined with historical data, a feature library for different ore characteristics is constructed to provide a basis for subsequent model reasoning. The extracted characteristic values are further processed using a recurrent neural network (RNN), and the predicted value is calculated through the trained model. The current grinding progress is evaluated based on the predicted value, and the future grinding trend is predicted. The output module converts the predicted value into an actual control signal, such as adjusting the speed of the ball mill, grinding time and other parameters. The control signal is output to the ball mill equipment control system in the form of a standard signal to realize the automatic control of the grinding process. Finally, the control effect feedback is collected, and the model is continuously optimized and adjusted to improve the prediction accuracy and control effect.
[0097] Furthermore, when the intelligent analysis module processes the acquired audio signal frequency band, it also includes:
[0098] Convert the acquired audio signal into a digital signal;
[0099] The converted digital signal is subjected to fast Fourier transform to convert the digital signal from the time domain to the frequency domain.
[0100] like Figure 3 As shown in the figure, by converting the signal from the time domain to the frequency domain through the fast Fourier transform (FFT), the frequency properties of the signal, such as high-frequency and low-frequency components and their bandwidth, can be viewed more intuitively. The spectrum diagram can clearly show the distribution of the signal in the frequency domain, which helps to identify and analyze the key features in the signal. In the frequency domain, FFT can separate the noise component in the signal and remove it through filtering and other methods, thereby improving the signal-to-noise ratio of the signal, which is particularly important in the monitoring of the ore grinding process, because the sound signals generated by different ores during grinding may have significant differences in the frequency domain.
[0101] Based on FFT, the system can also predict the spectral characteristics of future signals by analyzing historical spectrum data, so as to take corresponding control measures or optimization strategies in advance.
[0102] Specifically, the collected sound signal is filtered to remove noise and interference and improve signal quality. Subsequently, the signal is amplified for subsequent digital processing, and the ADC (analog-to-digital converter) converts the analog sound signal into a digital signal for processing. FFT (Fast Fourier Transform) analysis converts the digital signal from the time domain to the frequency domain, facilitating subsequent energy spectrum calculation.
[0103] Furthermore, when extracting the energy of each frequency band as a feature and combining the historical state to construct the feature extraction of the grinding progress of ores with different hardness / humidity, it also includes:
[0104] Divide the frequency domain signal into several fixed frequency band intervals and calculate the energy spectrum in each frequency band;
[0105] The calculated energy spectrum is used as subsequent input, and the model reasoning module uses a recurrent neural network to process the discrete energy spectrum data, extract the grinding progress characteristics, and predict the ball mill capacity.
[0106] By dividing the frequency domain signal into several fixed frequency bands and calculating the energy spectrum, the characteristics of the audio signal related to the grinding progress can be captured more accurately, which helps to improve the accuracy of the prediction model. Combining historical state data to build a feature set enables the model to capture dynamic changes in the grinding process, enhances the model's adaptability to different ore types and changes in grinding conditions, and improves the robustness of the model. Using RNN to process discrete energy spectrum data in real time, it is possible to achieve real-time monitoring and prediction of the grinding progress to optimize the grinding process.
[0107] Furthermore, the model reasoning module uses a recurrent neural network to process discrete energy spectrum data, extract grinding progress features, and predict ball mill capacity, specifically including:
[0108] First, Xt is input into the recurrent neural network as the input for subsequent calculations. The recurrent neural network introduces the historical input state Ct-1, processes the feature vector using the predefined weight matrix and activation function, and calculates the state Ht and intermediate state Ct of the current cycle.
[0109] like Figure 3 to Figure 4 As shown, specifically, in the frequency domain, the signal is divided into several fixed frequency intervals, and the energy spectrum in each frequency band is calculated, namely x0, x1, ..., xn. The energy spectrum data reflects the energy distribution of the sound signal in different frequency bands, which serves as the input for subsequent model reasoning. The model reasoning module mainly uses RNN to process discrete energy spectrum data. RNN captures the time dependency in the sequence data by introducing the historical input state Ct-1, and extracts features related to the grinding progress. The parallel processing capability of FPGA (field programmable gate array) is used to accelerate the calculation process of RNN and improve the system response time.
[0110] Further,
[0111] The activation function is Sigmoid; the intermediate state Ct=Sig(W1﹒Xt)+(W2﹒Ct-1);
[0112] The Ht=(Sig(W0﹒Xt)☉(Sig(Sig(W1﹒Xt)+(W2﹒Ct-1)));
[0113] Among them: Ht is the state of the current cycle; Sig is the activation function; W0, W1, W2 are predefined weight matrices; Xt is the input quantity; Ct-1 is the historical input state.
[0114] As shown in the figure, for example: take a three-dimensional vector as an example:
[0115] Xt=[0.5470 0.2963 0.7447]
[0116] W0=W1=[0.1890 0.6868 0.1835]T
[0117] W2=[0.0811 0.9294 0.7757]T
[0118] Ct-1=[0.3685 0.6256 0.7802]
[0119] W1﹒ Xt=W0﹒ Xt=[0.1034 0.0560 0.1407 0.3757 0.2035 0.5114 0.1004 0.0544 0.1367]
[0122] W2﹒ Ct-1=[0.0299 0.0508 0.0633 0.3425 0.5814 0.7251 0.2858 0.4853 0.6052]
[0125] Sig(W0﹒Xt)=Sig(W0﹒Xt)= [0.5258 0.5140 0.5351 0.5928 0.5507 0.6251 0.5251 0.5136 0.5341]
[0129] (Sig(Sig(W1﹒Xt)+(W2﹒Ct-1)))=[ 0.6355 0.6376 0.6453 0.7181 0.7562 0.7942 0.6923 0.7308 0.7576]
[0133] Ht=[1.1234 1.1195 1.2464]
[0134] Ct=Sig(W1﹒Xt)+(W2﹒Ct-1) can be used for subsequent data calculations
[0135] Ct = [0.5557 0.5647 0.5984 0.9353 1.1321 1.3503 0.8109 0.9989 1.1393]
[0138] Define the structure of the RNN, including the input layer, hidden layer (possibly multiple layers), and output layer.
[0139] Use the predefined weight matrix (such as W0, W1, W2) and activation function (such as Sigmoid) to perform forward propagation calculations to obtain the hidden layer state Ht and the final output Out.
[0140] Utilize the parallel processing capability of FPGA to accelerate the calculation process of RNN and improve the system response time.
[0141] Furthermore, the recurrent neural network is composed of multiple recurrent neural networks connected in series.
[0142] like Figure 5 As shown, the system can be composed of a multi-level RNN structure, the output is fully connected using Ht, Out is the final output, and the value is between [0,1]. If 4~20mA output is used, the final output is a current of 4+16*Out(mA);
[0143] For example, to get the full link layer parameters:
[0144] Wq=[0.04417 0.01967 0.04561 0.00133 0.00934 0.01017 0.08972 0.03074 0.09954]
[0147] Then Wk=[0.1629 0.0709 0.1867]
[0148] Final data output: Out = Sig (Sum (Wk)) = 0.6036
[0149] Output current: 13.6mA.
[0150] Where: Wq is the mapping weight used from the hidden state (or similar state) of the RNN to the fully connected layer input.
[0151] Furthermore, the output module converts the calculated predicted value into an actual control signal and outputs the calculation result as a standard signal, specifically including:
[0152] The output module outputs Ht according to the state of the current cycle of the recurrent neural network, calculates the final output value Out through the full connection layer, the value is between [0,1], and converts it into a 4~20mA current signal for final output;
[0153] Out=Sig(Sum(Wk));
[0154] Among them: Sig is the activation function, Wk is the final weighted vector before the output of the fully connected layer.
[0155] In the model inference module of the present application, RNN calculates the current hidden layer state Ht and the final output Out based on the input energy spectrum data and historical state information. Out is a value between [0,1], representing the predicted value of the ball milling capacity.
[0156] The output module calculates the final prediction result through the full connection layer according to the output value of the RNN. Then, this prediction result is converted into a 4-20mA current signal to interface with the industrial control system.
[0157] The standard signal is Modbus protocol, which is used to send the current signal to the control system or display device to achieve remote monitoring and adjustment. The final output is calculated using a fully connected network, and the Sigmoid function is used as the activation function. The system output can be directly mapped to 4-20mA by the activation function (or the ratio can be output through Modbus)
[0158] The entire device of this application realizes the complete process from sound signal acquisition, processing, model reasoning to output control through close cooperation and integration between various modules, and realizes real-time monitoring and prediction of the grinding progress of the ball mill through high-speed data processing and transmission capabilities. At the same time, through precise calculation and conversion, the accuracy and reliability of the prediction results are guaranteed.
[0159] The above description is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this patent can make some changes or modifications to equivalent embodiments of equivalent changes by using the technical content suggested above without departing from the scope of the technical solution of the present invention. The implementation scheme in the above embodiment can also be further combined or replaced. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the solution of the present invention.
Claims
1. A grinding sound analyzer, It is characterized in that include: A microphone is used to receive the audio signal frequency band during the operation of the ball mill; A filtering module is used to filter the acquired audio signal frequency band and select the required audio signal frequency band; An amplification module, used for amplifying the filtered audio signal frequency band; An intelligent analysis module is used to obtain the amplified audio signal frequency band, process and analyze the audio signal frequency band and output prediction; An output module is used to convert the output prediction result into an actual control signal and output it; The processing and analysis module includes: an energy spectrum calculation module, which performs spectrum analysis on the frequency band of the audio signal, extracts the energy of each frequency band as a feature, and constructs a feature vector in combination with the historical state; The model reasoning module uses the constructed feature vector to process and predict the current ball mill capacity.
2. A grinding sound analyzer according to claim 1, It is characterized in that The energy spectrum calculation module includes: a spectrum analysis unit, which is used to receive the frequency band of the audio signal and divide the audio signal into a number of fixed frequency band intervals; An energy spectrum calculation unit, used for calculating the spectrum integral value within each frequency band interval; The feature extraction unit is used to take the spectrum integral value of the energy spectrum calculation module as a feature vector, and combine it with the intermediate state quantity of the previous cycle, use linear transformation and activation function to process the feature vector, and construct the feature vector in combination with the historical state.
3. A grinding sound analyzer according to claim 2, It is characterized in that The model reasoning module includes: a recurrent neural network unit, which is used to calculate the state and intermediate state quantity of the current cycle by combining the current input and the intermediate state quantity of the previous cycle using the recurrent neural network unit's cyclic structure; the model reasoning module includes a multi-level recurrent neural network unit to form a fully connected layer structure; The prediction output unit takes the output of the recurrent neural network unit as the input of the fully connected layer structure, and obtains the final prediction value through linear transformation and activation function.
4. A control method for a ball mill analyzer according to any one of claims 1 to 3, It is characterized in that The control method comprises the following steps: The receiver receives the audio signal from the ball milling equipment, and filters the audio signal to remove noise and unnecessary interference; The intelligent analysis module processes the acquired audio signal frequency bands, extracts the energy of each frequency band as features, and extracts features of the grinding progress of ores with different hardness / humidity based on historical status. The model reasoning module uses a recurrent neural network to extract features from energy spectrum data and calculate prediction values; The output module converts the calculated predicted value into an actual control signal and outputs the calculation result as a standard signal.
5. A control method for a ball mill analyzer according to claim 4, It is characterized in that When the intelligent analysis module processes the acquired audio signal frequency band, it also includes: Convert the acquired audio signal into a digital signal; The converted digital signal is subjected to fast Fourier transform to convert the digital signal from the time domain to the frequency domain.
6. A control method for a ball mill analyzer according to claim 5, It is characterized in that When extracting the energy of each frequency band as a feature and combining the historical state to construct the grinding progress of ores with different hardness / humidity, the following features are also included: Divide the frequency domain signal into several fixed frequency band intervals and calculate the energy spectrum in each frequency band; The calculated energy spectrum is used as subsequent input, and the model reasoning module uses a recurrent neural network to process the discrete energy spectrum data, extract the grinding progress characteristics, and predict the ball mill capacity.
7. A control method for a ball mill analyzer according to claim 6, It is characterized in that The model inference module uses a recurrent neural network to process discrete energy spectrum data, extract grinding progress features, and predict ball mill capacity, specifically including: First, Xt is input into the recurrent neural network as the input for subsequent calculations. The recurrent neural network introduces the historical input state Ct-1, processes the feature vector using the predefined weight matrix and activation function, and calculates the state Ht and intermediate state Ct of the current cycle.
8. A control method for a ball mill analyzer according to claim 7, It is characterized in that The activation function is Sigmoid; the intermediate state Ct=Sig(W1﹒Xt)+(W2﹒Ct-1); The Ht=(Sig(W0﹒Xt)☉(Sig(Sig(W1﹒Xt)+(W2﹒Ct-1))); Among them: Ht is the state of the current cycle; Sig is the activation function; W0, W1, W2 are predefined weight matrices; Xt is the input quantity; Ct-1 is the historical input state.
9. The control method of a ball mill analyzer according to claim 4, It is characterized in that The recurrent neural network is composed of a plurality of recurrent neural networks connected in series.
10. The control method of a ball mill analyzer according to claim 7, It is characterized in that The output module converts the calculated predicted value into the actual control signal and outputs the calculation result as a standard signal. Specifically, it includes: The output module outputs Ht according to the state of the current cycle of the recurrent neural network, calculates the final output value Out through the full connection layer, the value is between [0,1], and converts it into a 4~20mA current signal for final output; Out=Sig(Sum(Wk)); Among them: Sig is the activation function, Wk is the final weighted vector before the output of the fully connected layer.
Citation Information
Patent Citations
Ball grinding mill load detecting method for optimizing BP neural network based on genetic algorithm
CN103839106A
Ball mill working condition identification method, system and device based on sound time-frequency domain analysis
CN111640427A
Voice processing method, device and equipment and computer readable storage medium
CN116682453A
Method and device for intelligently judging coal mine production state based on spectral analysis
CN117171655A
Intelligent grinding control system and method based on deep learning
CN117206063A
