Control system of an apparatus for drying metal powder containing combustible volatile substances

Abstract

A control system for an apparatus for drying metal powder containing combustible volatile substances belongs to the field of automatic control technology. In the system, a first subtractor generates a first sequential error signal based on the weights of two metal powders provided by a weighing sensor at time intervals; a second subtractor generates a second sequential error signal based on the measured temperature inside the drying chamber and a set temperature provided by a temperature sensor; a reinforcement learning module generates a first state and a second state based on the first and second sequential error signals, respectively, and generates a control strategy for a PID controller to control the operating state of the heater based on the first and second states; a fire detection model estimates the probability of a fire occurring inside the drying chamber based on the oxygen concentration, volatile substance concentration, and measured temperature. If the probability exceeds a threshold, stored inert gas is injected into the drying chamber to reduce the probability of a fire. This invention avoids the explosion risk caused by the accumulation of combustible gas.

Description

Technical Field

[0001] This invention relates to a control system for an apparatus for drying metal powder containing combustible volatile substances, belonging to the field of automatic control technology. Background Technology

[0002] Chinese invention patent application CN120215603A discloses a temperature and humidity adaptive control system for drying metal powder. It includes an environmental parameter comparison and analysis module, which analyzes the environmental parameters acquired by the drying environment parameter acquisition module, compares them with standard drying parameters to generate normal monitoring signals and parameter adjustment signals, and transmits the parameter adjustment signals to an adaptive adjustment and analysis module. The adaptive adjustment and analysis module analyzes the acquired parameter adjustment signals to generate temperature adjustment information, obtains a change coefficient by analyzing the relationship between temperature and humidity, and adjusts humidity using temperature as a standard comprehensive change coefficient to obtain a humidity value. Simultaneously, it compares the humidity value with standard drying parameters to generate humidity adjustment information and a secondary analysis signal, and transmits the secondary analysis signal to a secondary comprehensive adjustment and analysis module. The secondary comprehensive adjustment and analysis module processes the acquired secondary analysis signals, compares the humidity value with the standard humidity value to generate high humidity and low humidity signals, analyzes the high humidity signal by analyzing the fan speed to generate speed adjustment information, and analyzes the low humidity signal by analyzing the water vapor input to generate input adjustment information, and transmits both to a control and adjustment information output module. This invention solves the problem of the lack of precise environmental parameter monitoring and intelligent adjustment mechanisms, making it difficult to achieve precise control of the drying environment based on the characteristics of different metal powders.

[0003] However, when metal powder contains flammable volatile substances, improper control parameters can lead to an explosion risk, and there are no reported methods, systems, or devices for drying metal powder containing flammable volatile substances in the prior art. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the purpose of this invention is to provide a control system for an apparatus for drying metal powder containing combustible volatile substances, which features self-optimization and self-adaptation of drying speed; avoids the risk of explosion caused by the accumulation of combustible substances, and improves drying quality.

[0005] To achieve the aforementioned objective, this invention provides a control system for an apparatus for drying metal powder containing combustible volatile substances. The system includes a first concentration sensor, a second concentration sensor, a temperature sensor, a weighing sensor, a first subtractor, a second subtractor, a reinforcement learning module, and a fire detection model. The first subtractor generates a sequential first error signal based on the weights of two metal powders provided by the weighing sensor at time intervals. The second subtractor generates a sequence of second error signals based on the measured temperature inside the drying chamber provided by the temperature sensor and the set temperature. The reinforcement learning module uses the first error signal of the sequence. Generate the first state E 1t According to the second error signal of the sequence Generate the second state E 2t According to the first state E 1t Second state E 2t Generate the control strategy of the PID controller for controlling the operating state of the heater at time t. , It is the parameter vector of the PID controller. The heater is used to provide heat energy to the drying chamber. The fire detection model estimates the probability value of a fire in the drying chamber based on the concentration of combustible volatile substances in the drying chamber obtained by the first concentration sensor, the oxygen concentration in the drying chamber obtained by the second concentration sensor, and the measured temperature. If the probability value exceeds the threshold, the first solenoid valve opens, so that the storage container storing inert gas is connected to the drying chamber through the first solenoid valve.

[0006] Compared with the prior art, the control system of the device for drying metal powder containing combustible volatile substances provided by the present invention generates a control strategy for a PID controller that controls the working state of the heater through a reinforcement learning module, thereby enabling the drying speed to be self-optimized and adaptive, and improving the drying quality; and avoids the risk of explosion caused by the accumulation of combustible gas by controlling the working state of the first electrically controlled valve for injecting inert gas through a fire detection model. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the apparatus provided by the present invention for drying metal powder containing combustible volatile substances.

[0008] Figure 2 This is a block diagram of the control system of the apparatus for drying metal powder containing combustible volatile substances provided by the present invention.

[0009] Figure 3 This is a schematic diagram of the neural network composition of the reinforcement learning module provided by the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Figure 1This is a schematic diagram of the apparatus provided by the present invention for drying metal powder containing combustible volatile substances, as shown in the figure. Figure 1 As shown, the apparatus for drying metal powder containing combustible volatile substances provided by the present invention includes a drying chamber 1, a sensor module 2, a heater 3, and a control system 4. The heater 3 heats the drying chamber according to instructions from the control system. The drying chamber is used to hold the metal powder containing combustible volatile substances. The sensor module includes a first concentration sensor, a temperature sensor, and a weighing sensor. The probe of the first concentration sensor is disposed within the drying chamber to acquire the concentration of combustible volatile substances volatilized during the drying of the metal powder, and provides the measured concentration information of the combustible volatile substances to the control system 4. The probe of the temperature sensor is disposed within the drying chamber to acquire the temperature within the drying chamber, and provides the measured temperature to the control system 4. The weighing sensor measures the weight of the metal powder within the drying chamber and provides the weight information to the control system 4.

[0012] The apparatus for drying metal powder containing combustible volatile substances also includes an exhaust fan 8. When drying metal powder, the metal powder produces volatile toxic gases. The control system provides a control signal to the exhaust fan, which filters the toxic gases in the drying chamber 1 through the exhaust filter 9 and stores them in the toxic gas storage 7. At the same time, the control system provides a control signal to the second solenoid valve 10 to open the second solenoid valve, and outside air is injected into the drying chamber through the intake filter 11, thus forming continuous ventilation.

[0013] The sensor module also includes a second concentration sensor, which is used to acquire the oxygen concentration in the drying chamber and provide the concentration information to the control system. The device for drying metal powder containing combustible volatile substances also includes a first electrically controlled valve 6, which is installed on the pipeline connecting the storage container 5 storing inert gas to the drying chamber 1. The control system includes a fire detection model, which predicts the probability of fire based on the information provided by the first concentration sensor, the second concentration sensor, and the temperature sensor. When the probability of fire is greater than a threshold, it provides a control signal to the first electrically controlled valve and controls its opening and closing degree to inject inert gas into the drying chamber, thereby reducing the concentration of combustible volatile substances and oxygen concentration, and reducing the probability of fire.

[0014] Figure 2 This is a block diagram of the control system of the apparatus for drying metal powder containing combustible volatile substances provided by the present invention, as shown below. Figure 2 As shown, the control system includes: a first concentration sensor, a second concentration sensor, a temperature sensor, a weighing sensor, a first subtractor 21, a second subtractor 22, a reinforcement learning module, and a fire detection model. The first subtractor calculates the weight of the metal powder based on time intervals provided by the weighing sensor. and Generate the first error signal of the sequence In this invention, the weight of the metal powder obtained by the weighing sensor at time t is: The weight of the metal powder supplied to the first subtractor by the delay timer at time t-1 is... The second subtractor is based on the measured temperature inside the drying chamber provided by the temperature sensor. and set temperature Generate the second error signal of the sequence The reinforcement learning module uses the first error signal of the sequence. Generate the first state E 1t According to the second error signal of the sequence Generate the second state E 2t According to the first state E 1t Second state E 2t Generate the control strategy of the PID controller for controlling the operating state of the heater at time t. , It is the parameter vector of the PID controller; the heater is used to provide heat energy to the drying chamber; the fire detection model is based on the concentration of combustible volatile substances obtained by the first concentration sensor. The second concentration sensor obtains the oxygen concentration in the drying chamber. and measured temperature The probability of a fire occurring in the drying chamber is estimated. If the probability exceeds a threshold, the first electrically controlled valve 6 is opened, connecting the storage container for the inert gas to the drying chamber through the first electrically controlled valve to reduce the probability of a fire.

[0015] In this invention, , These are the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller, respectively.

[0016] Still Figure 2 As shown, the output of the PID controller is The set temperature of the drying chamber at time t and actual temperature The error is Then we have:

[0017] ,

[0018] In the formula, , ; .

[0019] Written in matrix form:

[0020] ,

[0021] In the formula,

[0022] , ,

[0023] In the formula, the time step is 1; T represents transpose.

[0024] However, the heater is also controlled based on the drying rate of the metal powder. Therefore, in this embodiment, the control parameters of the PID controller that controls the operating state of the heater are adjusted as follows:

[0025] ,

[0026] In the formula, These are the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller.

[0027] ,

[0028] In the formula, ; ; The time step is 1; T represents transpose.

[0029] The control system also includes a comparator for comparing the concentration of the combustible volatile substance acquired by the first concentration sensor. The concentration of combustible volatile substances is compared with the set concentration threshold. If it is greater than or equal to the threshold, a control signal is provided to the exhaust fan 8 to start the exhaust fan and extract the gas in the drying chamber. At the same time, the second solenoid valve 10 is opened, and the outside air enters the drying chamber through the opened second solenoid valve.

[0030] Figure 3 This is a schematic diagram of the neural network composition of the reinforcement learning module provided by the present invention, as shown below. Figure 3 As shown, the neural network of the reinforcement learning module includes an input layer, a deformable layer, a hidden layer, and an output layer. The input layer includes two neurons, which are respectively input to the first error signal of the sequence. and the second error signal of the sequence The deformable layer consists of 6 neurons, grouped into groups of 3. The first neuron group is based on the first error signal of the sequence. Generate the first state E 1t The second neuron group generates the second state E based on the second error signal of the sequence. 2t .

[0031] In this invention, the hidden layer contains 6 neurons; the output layer contains 5 neurons, wherein the first to third neurons of the output layer output the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller, respectively.

[0032] ,

[0033] In the formula, Let a be the output function of the a-th neuron in the hidden layer. and These are the center and width of the a-th Gaussian function in the neural network of the reinforcement learning module; These are the weights between the a-th neuron in the hidden layer and the m-th neuron in the output layer of the neural network for the reinforcement learning module; a=1,…,6; m=1,2,3. Represents the 2-norm; Indicates splicing.

[0034] In this invention, the 4th and 5th neurons of the output layer output the temperature state function of the drying chamber and the dryness state function of the metal powder, respectively:

[0035] ,

[0036] In the formula, To reinforce the learning model, we define the weights between the a-th neuron in the hidden layer and the M-th neuron in the output layer at time t, where a = 1, ..., 6; M = 4, 5.

[0037] In this invention, the reinforcement learning module also constructs a cost function according to the following formula:

[0038] ,

[0039] In the formula,

[0040] ,

[0041] In the formula, , These are the weighting coefficients;

[0042] ,

[0043] In the formula, It is a symbolic function; For coefficients; and These are the weights of the metal powder measured by the weighing sensor at time t and time t-1, respectively. The set temperature of the drying chamber at time t; The measured temperature of the drying chamber at time t is given.

[0044] In this invention, the reinforcement learning module updates the weights between the a-th neuron in the hidden layer and the m-th neuron in the output layer of the neural network at time t using gradient descent based on the cost function.

[0045] ,

[0046] -,

[0047] ,

[0048] ,

[0049] In the formula, , , , The learning coefficient; To strengthen the learning module, the weights between the a-th neuron in the hidden layer and the m-th neuron in the output layer at time t+1 are given, where a = 1, ..., 6. The weights between the a-th neuron in the hidden layer and the M-th neuron in the output layer of the neural network in the reinforcement learning module at time t+1, where M=4,5; To find the center of the a-th Gaussian function in the neural network of the reinforcement learning module at time t+1; The bandwidth of the a-th Gaussian function in the neural network for the reinforcement learning module at time t+1.

[0050] In the first embodiment, the reinforcement learning module also determines the cost function. Is it the smallest? If not, , , , Repeat the above steps. If yes, output... , , , As the optimal parameter for calculation , and make The proportional coefficient, integral coefficient, and derivative coefficient of the PID controller are assigned values ​​respectively.

[0051] Furthermore, although the parameters of the neural network in the reinforcement learning module are dynamically updated over time using gradient descent, the initial selection of these parameters is crucial for achieving the desired results.

[0052] In this invention, the fire detection model is trained from the following models:

[0053] ,

[0054] In the formula, These are the measured temperature inside the drying chamber, the measured concentration of combustible volatile substances, and the measured concentration of oxygen, respectively. These are model parameters; Let p be the probability, and p be the total probability value.

[0055] In this invention, the fire detection model training process includes:

[0056] S1-1: A series of data were obtained by using a digital twin model of an apparatus for simulating the drying of metal powder containing combustible volatile substances, forming a data vector. The n is a positive integer greater than or equal to 2.

[0057] In this invention, the fire detection model training process also includes:

[0058] S1-2: From data vector Obtain a set of positive samples:

[0059] q is a positive integer less than or equal to n;

[0060] S1-3: Perturb a set of positive samples to generate a set of perturbed samples:

[0061] ;

[0062] S1-4: Updated using the following formula :

[0063] ,

[0064] In the formula, The learning coefficient, For about The gradient;

[0065] S1-5: According to the updated A fire identification model was obtained.

[0066] The present invention can improve the generalization ability of the fire detection model through the above training method. In addition, although the present invention constructs the above fire detection model based on the measured temperature, measured concentration of combustible volatile substances and measured concentration of oxygen in the drying chamber, other variables can be added, such as the air pressure in the drying chamber. Various deformations of the fire detection model due to the addition of variables are also within the scope of the present invention.

[0067] The control system of the device for drying metal powder containing combustible volatile substances provided by the present invention generates a control strategy for a PID controller that controls the working state of the heater through a reinforcement learning module, thereby enabling the drying speed to be self-optimized and adaptive, and improving the drying quality; and avoids the risk of explosion caused by the accumulation of combustible gas by controlling the working state of the first electrically controlled valve for injecting inert gas through a fire detection model.

[0068] The preferred embodiments of the present invention disclosed above are only for the purpose of illustrating the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A control system for an apparatus for drying metal powder containing combustible volatile substances, characterized in that, include: The system comprises a first concentration sensor, a second concentration sensor, a temperature sensor, a weighing sensor, a first subtractor, a second subtractor, a reinforcement learning module, and a fire detection model. The first subtractor generates a sequence of first error signals based on the weights of the two metal powders provided by the weighing sensors at time intervals. ; The second subtractor generates a sequence of second error signals based on the measured temperature inside the drying chamber provided by the temperature sensor and the set temperature. ; The reinforcement learning module uses the first error signal of the sequence. Generate the first state E 1t According to the second error signal of the sequence Generate the second state E 2t According to the first state E 1t Second state E 2t Generate the control strategy of the PID controller for controlling the operating state of the heater at time t. , It is the parameter vector of the PID controller, and the heater is used to provide heat energy to the drying chamber; The fire detection model estimates the probability of a fire in the drying chamber based on the concentration of combustible volatile substances obtained by the first concentration sensor, the oxygen concentration obtained by the second concentration sensor, and the measured temperature. If the probability exceeds the threshold, the first electrically controlled valve opens, connecting the storage container for inert gas to the drying chamber through the first electrically controlled valve.

2. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 1, characterized in that, The neural network of the reinforcement learning module includes an input layer, a deformable layer, a hidden layer, and an output layer. The input layer consists of two neurons, which are respectively input to the first error signal of the sequence. and the second error signal of the sequence The deformable layer consists of 6 neurons, grouped into groups of 3. The first neuron group is based on the first error signal of the sequence. Generate the first state E 1t The second neuron group is based on the second error signal of the sequence. Generate the second state E 2t .

3. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 2, characterized in that, , , In the formula, the time step is 1.

4. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 3, characterized in that, The hidden layer contains 6 neurons; the output layer contains 5 neurons, where the first to third neurons of the output layer output the parameter vector A of the PID controller. 1t The three elements of a PID controller are the proportional coefficient, integral coefficient, and derivative coefficient: ， In the formula, Let a be the output function of neuron a in the hidden layer. and These are the center and width of the a-th Gaussian function in the neural network of the reinforcement learning module; These are the weights at time t between the a-th neuron in the hidden layer and the m-th neuron in the output layer of the neural network in the reinforcement learning module; a=1,…,6; m=1,2,3. Represents the 2-norm; Indicates splicing.

5. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 4, characterized in that, The 4th and 5th neurons in the output layer output the temperature state function of the drying chamber and the dryness state function of the metal powder, respectively: ， In the formula, To reinforce the learning model, we define the weights between the a-th neuron in the hidden layer and the M-th neuron in the output layer at time t, where a = 1, ..., 6; M = 4, 5.

6. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 5, characterized in that, The reinforcement learning module also constructs a cost function based on the following formula: ， In the formula, ， In the formula, , These are the weighting coefficients; , In the formula, It is a symbolic function; For coefficients; and These are the weights of the metal powder measured by the weighing sensor at time t and time t-1, respectively. The set temperature of the drying chamber at time t; The measured temperature of the drying chamber at time t is given.

7. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 6, characterized in that, The reinforcement learning module also updates the weights between the a-th neuron in the hidden layer and the m-th neuron in the output layer, as well as the weights between the a-th neuron in the hidden layer and the M-th neuron in the output layer, using gradient descent based on the cost function.

8. The control system of the apparatus for drying metal powder containing combustible volatile substances according to claim 1, characterized in that, The fire detection model was trained from the following models: ， In the formula, These are the measured temperature inside the drying chamber, the measured concentration of combustible volatile substances, and the measured concentration of oxygen, respectively. These are model parameters; It is a probability function; This represents the total probability value.

9. The control system of the device for drying metal powder containing combustible volatile substances according to claim 8, wherein the fire detection model training process includes: S1-1: A series of data were obtained by using a digital twin model of an apparatus for simulating the drying of metal powder containing combustible volatile substances, forming a data vector. The n is a positive integer greater than or equal to 2.

10. The control system of the device for drying metal powder containing combustible volatile substances according to claim 9, wherein the fire detection model training process further includes: S1-2: From data vector Obtain a set of positive samples: q is a positive integer less than or equal to n; S1-3: Perturb a set of positive samples to generate a set of perturbed samples: ； S1-4: Updated using the following formula : , In the formula, The learning coefficient, For about The gradient; S1-5: According to the updated A fire identification model was obtained.

Images