Rice steaming vehicle energy-saving control method based on pressure feedback

Abstract

The invention relates to the technical field of pressure feedback, and discloses a rice steaming vehicle energy-saving control method based on pressure feedback, pressure sensing is realized through heterogeneous layout of multiple types of piezoresistive, piezoelectric and capacitive sensors and a weighted fusion algorithm, and the detection precision is improved in combination with a temperature drift compensation mechanism; the dynamic power control module triggers mode switching according to pressure interval division and hysteresis adjustment, and power dynamic distribution is achieved through graded regulation and control of an enhanced set of electric heating tubes and pulse interval self-adaptive calculation. The leakage detection compensation module triggers graded power compensation based on voiceprint spectrum analysis, and synchronously links the deformation of the sealing mechanism to suppress leakage; the rice seed identification module adopts a high-resolution image acquisition and LBP-GLCM feature fusion technology, and is combined with an improved ResNet network to realize accurate rice seed identification, so that the problems of low pressure detection precision, leakage response lag and large energy consumption redundancy of a traditional rice steaming vehicle are solved, and the control precision of the steaming process and the operation safety of equipment are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure feedback, and particularly to an energy-saving control method for a rice steaming cabinet based on pressure feedback. Background Art

[0002] With the large-scale development of the catering industry, the automation and intelligence levels of commercial rice steaming cabinets have become the core factors affecting cooking efficiency and quality. As the direct control variable in the steaming process, the detection accuracy and real-time regulation of steam pressure directly determine the uniformity of rice taste, energy consumption economy, and equipment operation safety. Traditional rice steaming cabinet control systems usually rely on single-sensing technology to achieve pressure closed-loop regulation, but face multiple interference challenges such as steam flow field disturbance and temperature alternation under complex working conditions.

[0003] In the prior art, the pressure detection of rice steaming cabinets mostly uses a single type of sensor (such as piezoresistive or piezoelectric) combined with a fixed threshold control strategy. For example, some solutions collect pressure signals through piezoresistive sensors and adjust the power output of electric heating tubes based on the PID algorithm; other solutions use piezoelectric sensors to monitor dynamic pressure fluctuations and correct the measured values in combination with a temperature compensation circuit. In addition, there are also improved solutions in the prior art that improve the detection reliability by redundantly arranging multiple homogeneous sensors in hardware.

[0004] However, the above prior art has significant defects: the limited perception dimension of a single type of sensor makes it vulnerable to interference such as temperature drift and local flow field mutation, and it is difficult to accurately reflect the true pressure state of the entire domain of the steaming cabinet. For example, piezoresistive sensors generate zero drift due to thermal stress deformation in high-temperature environments, while piezoelectric sensors are sensitive to local steam turbulence and are prone to measurement value jumps. Although some solutions introduce a temperature compensation mechanism, the inherent physical characteristics of a single sensing source limit its ability to fundamentally solve the problem of inaccurate pressure detection under the coupling of multiple interferences, thereby leading to the failure of control strategies and an increased risk of steam leakage. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides an energy-saving control method for a rice steaming cabinet based on pressure feedback, which solves the problems of inaccurate pressure detection of the rice steaming cabinet caused by the limitations of a single sensor in the prior art and the resulting control failure and safety hazards.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: An energy-saving control method for a rice steaming cabinet based on pressure feedback, comprising the following steps:

[0007] Arrange piezoresistive, piezoelectric, and capacitive pressure sensors at the top of the cabinet, 30 cm from the bottom in the middle, and at the bottom;

[0008] Select a heating mode according to the interval distribution of the pressure value relative to the preset upper and lower thresholds, including overpressure cut-off mode, low-pressure full-power mode and transition zone proportional adjustment mode, where the upper and lower thresholds are dynamically adjusted according to the type of rice;

[0009] In the transition zone mode, the PWM duty ratios of the basic group and enhanced group electric heating tubes are dynamically allocated based on the pressure change rate calculated in a 1-second sliding window;

[0010] Synchronously collect the 32kHz sampling voiceprint signal of the dual microphone array and extract the energy proportion of the 500-1500Hz frequency band for leakage detection;

[0011] Control 12 groups of circumferentially arranged nickel-titanium alloy actuators, with a diameter of 1.2-1.5 mm, a phase transition temperature of 105-110°C, and a deformation of 0-0.3 mm;

[0012] The rice type was determined by classifying 60x microscopic images using the ResNet-18 network;

[0013] Predict pressure decay trends based on LSTM networks and trigger level 3 safety responses;

[0014] The TD3 reinforcement learning algorithm is used to optimize the control parameters, and the reward function weight ratio is energy efficiency 5: quality 3: loss 2.

[0015] Preferably, the sensor signal processing includes:

[0016] The output of the piezoresistive sensor is temperature compensated with a compensation coefficient of -0.15% / ℃;

[0017] The piezoelectric sensor signal is processed by a charge amplifier and a 4th-order Butterworth filter;

[0018] The capacitive sensor uses a differential measurement circuit with an excitation frequency of 1MHz;

[0019] The weighted fusion formula is:

[0020] P fusion =0.4P 压阻 +0.3P 压电 +0.3P 电容 .

[0021] Preferably, the dynamic allocation of electric heating pipe groups includes:

[0022] The electric heating tubes are divided into a basic group of 3kW×2 and an enhanced group of 6kW×1;

[0023] When the absolute value of the pressure change rate is greater than 5 kPa / s, the enhancement group is started first and works at 80% duty cycle PWM;

[0024] In the last 10 kPa stage when the pressure recovers to the safe range, the basic group switches to the alternating pulse mode, and the pulse interval is inversely proportional to the pressure deviation.

[0025] Preferably, the voiceprint feature correction includes: extracting the energy proportion of the 500 - 1500 Hz frequency band in the steam flow noise. When this proportion is lower than the preset value, it is determined that steam leakage has occurred, and the heating power compensation and sealing structure strengthening control are triggered.

[0026] Preferably, the spectrum feature correction includes:

[0027] Collecting voiceprint signals with a sampling rate of 32 kHz through a dual - microphone array;

[0028] Extracting the energy proportion value of the 500 - 1500 Hz frequency band. When this value is lower than 45%, it is determined as steam leakage;

[0029] Triggering the heating power compensation strategy: increasing the current power output by 15% - 25%, and the compensation duration is positively correlated with the leakage degree, with a maximum of no more than 120 seconds.

[0030] Preferably, the determination of the optimal pressure parameter combination includes:

[0031] Collecting 60 - fold microscopic images of the rice grain surface and extracting local binary pattern features and gray - level co - occurrence matrix features;

[0032] Outputting the rice variety type through a convolutional neural network classifier and associating with the preset pressure parameters:

[0033] Japonica rice: upper limit threshold 118 ± 2 kPa, dead - band width 12 kPa

[0034] Indica rice: upper limit threshold 125 ± 3 kPa, dead - band width 8 kPa

[0035] Glutinous rice: upper limit threshold 110 ± 1 kPa, dead - band width 15 kPa.

[0036] Preferably, the hierarchical safety mechanism includes:

[0037] First - level response: When the pressure decay rate exceeds 0.8 kPa / min, an audible and visual alarm is triggered and the fault log is recorded;

[0038] Second - level response: After the abnormal state lasts for 25 seconds, 50% of the heating power is cut off and the auxiliary pressure - relief valve is started;

[0039] Third - level response: When the local temperature is detected to exceed 130 °C, inert gas is released and an emergency shutdown signal is uploaded.

[0040] Preferably, the reinforcement learning algorithm includes:

[0041] Define the state space as a four-dimensional vector and the action space as a three-dimensional parameter adjustment amount;

[0042] The reward function is designed as a weighted combination of the energy efficiency factor, the cooking quality score, and the equipment loss coefficient, with a weight ratio of 5:3:2;

[0043] The policy update period is to perform batch learning once every 20 cooking tasks are completed.

[0044] Preferably, the method for enabling the standby redundant channel includes:

[0045] Disconnect the signal link of the suspected faulty sensor and inject a standard test pressure signal into the data bus;

[0046] If the system pressure reading fluctuates by more than ±3 kPa within 10 seconds, lock the sensor and switch to the adjacent sensor weighting mode;

[0047] Perform automatic zero calibration at the first startup every day, and maintain a reference pressure of 80 kPa during calibration.

[0048] Preferably, the control parameters of the nitinol actuator include:

[0049] Drive voltage 12 VDC, pulse width modulation frequency 8 kHz;

[0050] Temperature feedback sampling period 100 ms, PID control parameters K p =2.5, K i =0.02, K d =0.8;

[0051] The deformation self-check program is executed after each cooking, and the detection accuracy reaches 0.005 mm.

[0052] The present invention provides an energy-saving control method for a rice cooker based on pressure feedback. It has the following beneficial effects:

[0053] 1. The present invention adopts a heterogeneous layout scheme of piezoresistive, piezoelectric, and capacitive multi-type sensors, combined with a temperature drift compensation and weighted fusion algorithm, to achieve full-condition coverage of pressure detection and enhanced anti-interference ability. Compared with the traditional single-sensor scheme, it solves the problem of misjudgment caused by susceptibility to local flow field disturbance and temperature sensitivity, and significantly improves the robustness of pressure perception.

[0054] 2. The present invention triggers the hierarchical regulation of the enhanced group of electric heating tubes through the real-time feedback of the pressure change rate, combined with the pulse interval adaptive calculation technology, to achieve the dynamic matching of the heating power and the steam demand. Compared with the existing fixed threshold control method, it effectively avoids heating lag and power redundancy, and reduces the ineffective energy consumption.

[0055] 3. The present invention triggers a hierarchical compensation strategy based on voiceprint frequency band energy analysis, synchronously linking the deformation of the sealing mechanism and the power improvement. Compared with the traditional single-threshold alarm mechanism, it solves the defects of slow response and insufficient compensation strength in the initial stage of leakage, and realizes the fast closed-loop control of leakage suppression.

[0056] 4. The present invention introduces the state space modeling and multi-objective reward mechanism of reinforcement learning, making the control parameters have working condition self-adaptability. Compared with the manual parameter tuning scheme relying on experience, it breaks through the generalization ability bottleneck in complex scenarios and significantly improves the dynamic optimization level of the control strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic flow chart of the method of the present invention;

[0058] Figure 2 It is a schematic diagram of the device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0060] Please refer to the attached Figure 1-2 , the embodiment of the present invention provides an energy-saving control method for a rice cooker based on pressure feedback, including:

[0061] Sensor layout

[0062] In some embodiments, the installation position of the piezoresistive sensor (MPX5700AP) needs to meet geometric constraint conditions. The top installation point is 5.0 ± 0.2 cm away from the inner wall edge of the box, and the installation plane forms an angle ≤ 0.5° with the horizontal plane, and is calibrated and positioned by a laser level. The sensor cavity is filled with silicone gel (viscosity 5000 cP) to buffer mechanical vibration, and its attenuation coefficient is calculated as:

[0063]

[0064] Among them, A1 and A2 are the amplitude values before and after the vibration wave is introduced, respectively, and are obtained by actual measurement with an accelerometer (model PCB-352C33).

[0065] In some embodiments, the installation inclination angle of the piezoelectric sensor (PCB-113B28) is determined by trigonometric function relationships:

[0066]

[0067] In the formula, h offset = 3.2 cm is the vertical offset of the mounting bracket, and L mount = 3.2 cm is the horizontal fixing length to ensure that the shear force detection direction is consistent with the main direction of the steam flow field.

[0068] The installation position of the capacitive sensor (MS5803 - 05BA) can be configured with a porous flow deflector. The opening diameter of the flow deflector is 1.0 ± 0.05 mm, and the porosity is 35%. The calculation of its pressure drop loss is as follows:

[0069]

[0070] Among them, ρ = 1.127 kg / m 3 is the steam density, v = 2.5 m / s is the flow velocity, and C d = 0.62 is the orifice flow coefficient.

[0071] Grouping of electric heating tubes:

[0072] In some embodiments, the spiral winding parameters of the basic group of electric heating tubes need to meet the requirements of thermal field uniformity. The spiral diameter D = 12 ± 0.5 cm, the pitch S = 5.0 ± 0.2 cm, and the total length is calculated as:

[0073]

[0074] The thickness of the surface magnesium oxide layer is 0.8 ± 0.05 mm, the breakdown voltage > 3 kV, and the thermal conductivity is 1.3 W / (m·K). In some embodiments, the linear arrangement of the enhanced group of electric heating tubes needs to be configured with a thermal expansion compensation structure. The sheath material is selected as 310S stainless steel, and the linear expansion coefficient is 16.0×10 -6 / ℃, and the reserved expansion gap:

[0075] δ = αLΔT = 16×10 -6 ×1200×300 = 5.76 mm;

[0076] Among them, L = 1200 mm is the length of the electric heating tube, and ΔT = 300℃ is the working temperature rise. In alternative embodiments, the PWM carrier frequency error of the electric heating tube drive circuit needs to be controlled within ±50 ppm. A temperature-compensated crystal oscillator (TCXO, model ECS - 3225S) is used, and the frequency stability is calculated as:

[0077] Δf = f0×(α(T - T0) + β(T - T0) 2 );

[0078] Among them, f0 = 1 MHz, α = 0.5 ppm / ℃, β = 0.05 ppm / ℃2, and T0 = 25℃.

[0079] In some embodiments of the voiceprint acquisition module, the installation geometry of the microphone array needs to meet the acoustic interference condition. The maximum aliasing-free frequency corresponding to an axial spacing of 80 mm is:

[0080]

[0081] where c = 343 m / s is the speed of sound, which is higher than the target frequency band of 1500 Hz to ensure no phase ambiguity. In some embodiments, the transfer function of the band-pass filter is designed as:

[0082]

[0083] where ω0 = 2π×1000 rad / s, the quality factor Q = 2.5, and the stopband attenuation > 40 dB / dec. In an alternative embodiment, the integral nonlinear error of the ADC needs to be dynamically corrected. The two-point correction method is used:

[0084]

[0085] In the formula, V zero is the zero-point voltage (measured with respect to ground), V ref = 2.5 V is the reference voltage, and FS = 5 V is the full scale.

[0086] In this embodiment, the pressure fusion algorithm is based on a multi-sensor data correction and optimization fusion strategy, and through temperature drift compensation, signal conditioning, and weighted fusion processing, the accuracy of dynamic pressure perception is improved. The following is a detailed description of temperature compensation, signal processing, and weighted fusion:

[0087] Temperature compensation:

[0088] In some embodiments, polynomial fitting algorithm is used for temperature drift compensation of the piezoresistive sensor (MPX5700AP). The compensation coefficient α = -0.15% / °C is obtained through calibration tests: apply a standard pressure of 100 kPa in a temperature chamber of 20 - 120 °C, collect the temperature-output curve, and calculate:

[0089] P comp = P raw ×[1 + α(T current - T calib )];

[0090] where T calib = 25 °C is the factory calibration temperature, T current is the real-time temperature (collected by the PT100 sensor), and the compensation residual < ±0.2% FS.

[0091] In some embodiments, cubic spline interpolation method is used to verify the temperature drift error. Calibrate at three points of -10 °C, 25 °C, and 60 °C to construct a piecewise compensation function:

[0092]

[0093] The coefficient a1 = 3.2×10 -5 , b1 = -1.8×10 -3 , c1 = 0.12 is determined by least squares fitting.

[0094] Signal processing

[0095] In some embodiments, the signal conditioning of the piezoelectric sensor (PCB-113B28) uses a charge-voltage conversion circuit. The gain of the charge amplifier is set to 1V / pC, and the feedback capacitor C f = 100pF, and its transfer function is:

[0096]

[0097] where d 33 = 325pC / N is the piezoelectric constant, and F is the applied force. The cut-off frequency of the subsequent 4th-order Butterworth filter is 500Hz, and the transfer function is:

[0098]

[0099] ω0 = 2π×500rad / s, and the group delay compensation is 1.2ms.

[0100] In some embodiments, the capacitive sensor (MS5803-05BA) uses a differential measurement mode. The excitation signal is:

[0101] V exc (t) = Asin(2πf c t) + Asin(2π(f c +Δf)t);

[0102] A = 2.5V, f c = 1MHz, Δf = 10kHz, and the phase of the difference frequency signal is extracted by a lock-in amplifier:

[0103]

[0104] where I and Q are the in-phase / quadrature components, the capacitance change ΔC = kφ, and the proportionality coefficient k = 0.05pF / rad.

[0105] Weighted fusion:

[0106] The calculation of the weight coefficient is based on the calibration test data. The three types of sensors are repeatedly tested 30 times at 100kPa pressure, and the standard deviation of the error is calculated:

[0107]

[0108] It is measured that σ1 = 0.8 kPa, σ2 = 1.2 kPa, σ3 = 1.1 kPa, and the weight distribution formula is:

[0109]

[0110] Substituting the values, we get w1 = 0.396, w2 = 0.302, w3 = 0.302, and the final fusion value:

[0111] P fusion = 0.396P1 + 0.302P2 + 0.302P3;

[0112] In some embodiments, the weight coefficient dynamic adjustment strategy is as follows: when the output deviation of a single sensor exceeds 3σ i its weight is temporarily reduced to 0.5 times, and the weights of other sensors increase proportionally.

[0113] In this embodiment, the dynamic power control strategy is based on real-time state assessment of pressure feedback, and through multi-mode switching logic, enhanced group trigger mechanism, and pulse interval adaptive adjustment, precise dynamic distribution of heating power is achieved. The following is a detailed description of mode switching, enhanced group trigger, and pulse interval calculation:

[0114] Mode switching:

[0115] In some embodiments, the heating mode switching is performed according to the pressure value interval distribution. The pressure intervals are divided into:

[0116] Low-pressure area: P < P min (P min = 85 kPa).

[0117] Transition area: P min ≤ P ≤ P max (P max = 125 kPa).

[0118] Overpressure area: P > P max .

[0119] The hysteresis width H is dynamically adjusted according to the type of rice seeds, and the calculation formula is:

[0120]

[0121] The mode switching trigger condition is:

[0122] Entering the low-pressure area: P ≤ P min + H / 2 and lasting for 3 seconds. Entering the overpressure area: P ≥ P max - H / 2 and lasting for 2 seconds.

[0123] In an alternative embodiment, the hysteresis width can be extended to H = 10 - 18 kPa and dynamically adjusted by predicting the pressure trend through an LSTM network.

[0124] In some embodiments triggered by the enhancement group, the triggering condition of the enhancement group (6 kW electric heating tube) is determined based on the pressure change rate. The pressure change rate is calculated using a 1-second sliding window linear fitting:

[0125]

[0126] where t i is the timestamp (at 0.1-second intervals), and P i is the fused pressure value. When the following condition is met:

[0127]

[0128] the enhancement group is immediately activated, with the duty cycle set to 80% PWM and the pulse width fixed at 10 ms to avoid frequent on-off cycles. In some embodiments, the power output of the enhancement group is limited by temperature. The surface temperature T of the electric heating tube surface is monitored in real time. When:

[0129]

[0130] a derating protection is triggered, and the duty cycle decays linearly:

[0131]

[0132] where D0 = 80% is the initial duty cycle and t0 is the protection trigger time.

[0133] Pulse interval calculation:

[0134] In some embodiments, the basic group (3 kW × 2) switches to an alternating pulse mode at the end of the transition zone (P ≥ P max - 10 kPa). The pulse interval t interval is inversely proportional to the pressure deviation ΔP = |P - P set |:

[0135]

[0136] When ΔP > 20 kPa, the lower limit of t interval is locked at 0.5 seconds to avoid overshoot. In some embodiments, the pulse width is adjusted synchronously:

[0137] t width = min(2.0, 0.3ΔP + 0.5);

[0138] The phase difference of the pulse sequence is set to 180°, and the two groups of electric heating tubes are alternately turned on, improving the thermal field uniformity by 23%.

[0139] In this embodiment, the leakage detection compensation mechanism is based on the dynamic analysis of voiceprint spectrum features and the power closed-loop regulation strategy, and realizes the real-time suppression of steam leakage through the monitoring of the energy ratio in a specific frequency band and the multi-parameter linkage compensation. The following is a detailed description of the frequency band energy calculation and the power compensation strategy:

[0140] Frequency band energy calculation

[0141] In some embodiments, the voiceprint signal preprocessing adopts anti-aliasing filtering and window function weighting. The 32 kHz original signal collected by the dual microphone array (model INMP441) is processed by a 4th-order Butterworth low-pass filter (cutoff frequency 16 kHz) to eliminate high-frequency noise interference. The filtered signal is windowed with a Hanning window:

[0142]

[0143] After windowing, a 1024-point FFT is performed, and the frequency resolution is 31.25 Hz. The corresponding frequency point indices k of the target frequency band 500 - 1500 Hz are from k = 16 to k = 48, and the energy ratio calculation formula is:

[0144]

[0145] where X[k] is the complex spectrum value of the k-th frequency point. When E ratio < 45% within three consecutive sampling periods (600 ms), the leakage determination condition is triggered.

[0146] In some embodiments, the spectrum leakage suppression adopts the overlapping sampling method. The overlapping rate of each frame of data is 50% (512 points), and the total number of frames is doubled, reducing the standard deviation of the spectrum fluctuation to ±1.2%.

[0147] Power compensation strategy

[0148] In some embodiments, after the leakage is confirmed, hierarchical power compensation is performed. The calculation formula for the basic compensation power increment is:

[0149]

[0150] where the square root function is used to prevent overcompensation in case of small deviations. The power output correction formula:

[0151] P new = P base × (1 + ΔP comp );

[0152] The compensation duration is non-linearly correlated with the pressure deviation:

[0153]

[0154] For example, when |ΔP| = 8 kPa, the compensation lasts for 80 seconds; if |ΔP| = 15 kPa, the compensation duration increases to 125 seconds.

[0155] In some embodiments, the seal compensation mechanism is activated synchronously. The calculation of the incremental deformation of the nickel-titanium alloy actuator is as follows:

[0156] Δd = 0.03×|ΔP| + 0.1 unit: mm

[0157] The driving voltage is dynamically adjusted according to the deformation rate:

[0158]

[0159] Among them, the hyperbolic tangent function restricts the voltage overshoot, and the maximum driving voltage does not exceed 14 V.

[0160] In this embodiment, the rice variety identification module is based on multi-modal feature fusion and deep neural network technology. Through high-resolution image acquisition, texture feature extraction, and convolutional network training, accurate identification of rice variety types is achieved. The following is a detailed description of image acquisition, feature extraction, and network training:

[0161] Image Acquisition

[0162] In some embodiments, the image acquisition device is configured with a 60x optical microscope (Olympus DSX1000), the numerical aperture of the objective lens NA = 0.7, and the working distance is 12.5 mm. The color temperature of the annular LED light source is 5500 K, the illuminance adjustable range is 500 - 1500 lux, and the brightness fluctuation is controlled by PWM dimming <±3%. The rice grain samples are laid flat on the black stage, 20 grains are collected at a time, the image resolution is 2560×1920 pixels, and they are stored in 12-bit RAW format.

[0163] In some embodiments, the image preprocessing performs automatic calibration: 1. White balance correction: Calculate the gain coefficient based on a standard white board (reflectivity 98%):

[0164]

[0165] 2. Distortion correction: Use a checkerboard calibration board to calculate the radial distortion coefficients k1 = -0.12 and k2 = 0.03, and the correction formula:

[0166] x corr = x(1 + k1r 2 + k2r 4 ) ;

[0167] Among them, is the normalized image coordinate.

[0168] Feature Extraction

[0169] In some embodiments, the local binary pattern (LBP) feature adopts an improved algorithm for circular neighborhood. For the pixel point (x c , y c ), the LBP value is calculated as follows:

[0170]

[0171] The parameters are P = 8, R = 1.5 pixels, and the coordinates of the neighborhood points are calculated by bilinear interpolation:

[0172] x p = x c + Rcos(2πp / P), y p = y c + Rsin(2πp / P);

[0173] In some embodiments, for the gray-level co-occurrence matrix (GLCM) feature extraction, the directions are θ = 0°, 45°, 90°, 135°, and the distance d = 1 pixel. The calculation formula for contrast is:

[0174]

[0175] Calculation of energy and correlation:

[0176]

[0177] Network training

[0178] In some embodiments, the input layer of the improved ResNet-18 network is adjusted to a single-channel grayscale image, the size of the first-layer convolution kernel is modified to 5×5, and the stride is 2. The output dimension of the final fully connected layer is set to 3 (japonica / indica / glutinous rice). The loss function adopts weighted cross-entropy:

[0179]

[0180] The weight coefficient compensates for the problem of insufficient indica rice sample size.

[0181] In some embodiments, the optimizer adopts AdamW, the initial learning rate n = 3e-4, and the weight decay λ = 0.05. Learning rate scheduling strategy:

[0182]

[0183] It decays by 5% every 10 epochs and is trained for 200 rounds in total. Data augmentation includes random rotation of ±15°, Gaussian noise (σ = 0.01), and brightness jitter of ±10%.

[0184] Primary alarm

[0185] In some embodiments, the alarm trigger condition fuses the pressure anomaly and the leakage detection result. When the pressure deviates from the set value by more than the threshold and remains invalid

[0186] When adjusting, the trigger conditions are:

[0187] (|ΔP|≥25kPa and lasts t≥15s) or (E ratio <40% and ΔP>18kPa) sound and light alarm (model AD16-22SM) starts three long and two short pulse signals, the buzzer frequency is f buzz =2Hz, LED red light flashes every 0.5 seconds. The alarm signal is uploaded to the central controller via RS485, and the response delay is <200ms.

[0188] In some embodiments, the alarm release conditions must be met at the same time:

[0189] |ΔP|≤8kPa and E ratio ≥48% for t≥30s

[0190] In an alternative embodiment, a vibration alarm module may be added, with the acceleration threshold set to 0.5g, and synchronous triggering when the frequency band energy (20-100Hz)>0.1V{}2.

[0191] Secondary power failure

[0192] In some embodiments, the power-off logic is based on the temperature-current dual parameter determination. surface >365℃ or the current mutation rate meets:

[0193]

[0194] The main relay (model G7L-2A-TUB) immediately cuts off the power supply, and the opening time is less than 50ms. After the power is cut off, the residual charge is discharged, the discharge resistance R = 10kΩ, and the voltage decays to a safe value V safe ≤24V time is calculated as:

[0195] t discharge =5τ=5RC=5×0.1×10 4 =5s;

[0196] Among them, C = 100μF is the bus capacitance.

[0197] In some embodiments, the power-off state latch needs to be manually reset. The reset conditions include:

[0198] The temperature drops to T surface ≤45℃

[0199] The pressure is restored to P∈[85,125]kPa

[0200] Manually press the reset button for 3 seconds.

[0201] Three-level fire extinguishing:

[0202] In some embodiments, the fire extinguishing trigger is determined based on the combination of an infrared flame sensor (model TSL261R) and a temperature gradient. The flame signal threshold V flame > 2.5V and the temperature change rate at the top of the cabinet satisfies:

[0203]

[0204] The release amount of the fire extinguishing agent (heptafluoropropane) is calculated as:

[0205]

[0206] where V chamber = 1.2 m 3 is the volume of the cabinet, A floor = 0.8 m 2 is the bottom area, and T max is the maximum temperature rise.

[0207] In some embodiments, the fire extinguishing agent is ejected by nitrogen drive, and the drive pressure P drive = 4.0 ± 0.2 MPa, and the nozzle flow rate is calculated as:

[0208]

[0209] When P atm = 101.325 kPa, v = 235 m / s, and the coverage radius is 1.5 m.

[0210] In this embodiment, the reinforcement learning optimization module is based on the Deep Deterministic Policy Gradient (DDPG) framework, and realizes the adaptive optimization of the control parameters of the rice cooker through multi-dimensional state space modeling, multi-objective reward function design, and dynamic update of the policy network. The following details the state space definition, reward function construction, and policy update mechanism: State space In some embodiments, the state vector s t contains the normalized multi-source sensing data and historical control quantities:

[0211]

[0212] where P current ∈ [85, 125] kPa is the current pressure value, and h t-3:t is the stacking of the historical states of the previous 3 steps. The time series data is filtered by moving average:

[0213]

[0214] In an alternative embodiment, the state space can be extended to include rice variety encoding (japonica rice 0 / indica rice 1 / glutinous rice 2), and is embedded in dimensions using One-hot encoding. Reward function In some embodiments, the reward function rt Integrate pressure tracking accuracy, energy consumption efficiency, and safety constraints:

[0215]

[0216] Among them, is an indicator function that takes 1 when the condition is met and 0 otherwise. The coefficient 10 of the pressure deviation term is determined by grid search, and the quadratic penalty of the energy consumption term suppresses power overshoot. In some embodiments, a sparse reward enhancement mechanism is introduced. When |ΔP| ≤ 5 kPa and T steam ∈ [98, 102] °C are satisfied for 10 consecutive steps, an additional reward is given:

[0217]

[0218] Policy update: In some embodiments, the Actor network uses three-layer fully connected (256-128-64), and the output layer uses Tanh activation to constrain the action within [-1, 1]. The action vector a t is mapped to the actual control quantity:

[0219] ΔP set = 5a1, ΔD base = 10a2, ΔD boost = 15a3

[0220] The input of the Critic network outputs the Q-value estimate. The soft update coefficient of the target network τ = 0.005, and the discount factor γ = 0.95. In some embodiments, the PPO-Clip algorithm is used for policy optimization. The importance sampling ratio is calculated as:

[0221]

[0222] The objective function limits the policy update amplitude:

[0223]

[0224] Among them, ∈ = 0.2, and the advantage function is estimated by GAE (λ = 0.9).

[0225] Furthermore, the specific liquid level linkage water replenishment control is as follows:

[0226] Installation location: Vertically installed 5 cm from the bottom of the inner liner of the rice steaming box. The effective measurement range of the probe is 0 - 30 cm, covering a water depth of 25 cm (the total length of the probe is 30 cm, and the bottom 5 cm is a fixed base)

[0227] Sensor model: Capacitive liquid level sensor (EASYDELL ED-688), output signal 4 - 20 mA, resolution ±1 mm

[0228] Probe material: 316L stainless steel shell, PTFE insulation layer (temperature resistant 200℃)

[0229] Sealing level: IP67, withstand continuous operation in steam environment > 5000 hours

[0230] Calibration mechanism: Two-point calibration of empty tank (0cm) and full tank (30cm) is preset before leaving the factory. The user can trigger the automatic calibration procedure (press and hold the "Calibration" button on the operation panel for 5 seconds).

[0231] Solenoid valve control:

[0232] Liquid level trigger conditions:

[0233] Model: SMC VDW21 (normally closed), response time ≤ 0.3s, flow rate 1.2L / min (@0.2MPa water pressure);

[0234] Drive circuit: ULN2003 Darlington array drive, relay contact capacity 5A / 250VAC;

[0235]

[0236] Anti-interference:

[0237] The liquid level sampling frequency is 10Hz, and the median filter algorithm (window width 15) is used;

[0238] When the steam pressure is >0.08MPa, the water replenishment operation is locked to prevent false triggering;

[0239] Dry-boiling protection and system recovery:

[0240] Protection action sequence:

[0241] When the liquid level is less than 5cm, the controller cuts off the power supply of the electric heating tube within 0.5s (the relay release time is less than 50ms);

[0242] The 85dB sound and light alarm (ABT-85S) is triggered simultaneously, and the fault code E14 pops up on the display screen;

[0243] Recovery conditions:

[0244] After manually confirming the fault, add water until H≥8cm and remain stable for 10s, and press the "Reset" button to release the alarm;

[0245] The system automatically records the fault time and recovery log (stored in FLASH chip W25Q128);

[0246] Pressure threshold control:

[0247] Pressure relief valve parameters:

[0248] Model: DN15 Stainless Steel Electromagnetic Pressure Relief Valve (pressure resistance 0.15MPa, leakage class IV);

[0249] Driving method: 24V DC electromagnetic coil, power consumption 8W, response time ≤ 0.5s;

[0250] Pressure relief diameter: 15mm, measured pressure relief rate 0.02 - 0.025MPa / s (@0.1MPa initial pressure);

[0251] Control conditions:

[0252] Opening condition: P ≥ 0.1MPa, P ≥ 0.1MPa for 5 seconds continuously (to prevent false triggering caused by instantaneous fluctuations);

[0253] Closing condition: P ≤ 0.095MPa, P ≤ 0.095MPa or pressure relief timeout (forced closing within 30 seconds at most);

[0254] Safety redundancy: Cross - verification by dual pressure sensors, trigger manual inspection alarm when the difference between the readings of the two sensors > 3kPa;

[0255] Hierarchical control of electric heating tubes:

[0256] Configuration of heating tubes:

[0257] Basic group: 3kW × 2 pieces (in parallel, maximum current 13.6A);

[0258] Enhanced group: 6kW × 1 piece (controlled by independent relay);

[0259] New emergency group: 4kW × 1 piece (dedicated for customer needs, PID power regulation module);

[0260] Interval control strategy:

[0261]

[0262] Pulse algorithm:

[0263]

[0264] Example: When P = 0.095MPa, single - time heating duration = 15s, intermittent duration = 15s

[0265] Ripening guarantee mechanism:

[0266] Time - pressure coupling constraint:

[0267] Accumulative heating time within 30 minutes ≥ 18 minutes (duty cycle ≥ 60%);

[0268] Pressure fluctuation limit: |ΔP| ≤ 3kPa (through PID parameter K p= 1.2, K i = 0.05, K d = 0.1 implementation);

[0269] Verification of rice grain ripening:

[0270] Test method: Glutinous rice samples (with a moisture content of 14%) are steamed under a pressure fluctuation of 0.09 - 0.1 MPa

[0271] Result: The gelatinization degree of the central rice grains reaches 92% (88% for traditional constant - pressure steaming), and the moisture content gradient difference < 2%;

[0272] Hardware configuration of multi - modal sensors

[0273] Spatial layout and installation specifications:

[0274] Top sensor: Capacitive pressure sensor (model Honeywell HSC series), installed at the center of the top of the steaming box, 2 cm from the top wall, measurement range 0 - 0.15 MPa, accuracy 0.5% FS, temperature resistance 150 °C.

[0275] Middle sensor: Piezoelectric dynamic pressure sensor (PCB Piezotronics 113B28), located at three equally - spaced points circumferentially at a height of 30 cm from the bottom, resonant frequency 1 kHz, sensitivity 50 mV / kPa, resistant to steam impact.

[0276] Bottom sensor: Piezoresistive sensor (Sensata P51), embedded in the bottom heating layer of the steaming box, temperature compensation range 0 - 150 °C, linear error < 0.1%, protection level IP69K.

[0277] Signal acquisition chain:

[0278] Piezoresistive signal: Sampled by a 24 - bit ADC (ADS1220), and the temperature compensation circuit is integrated on the PCB board (compensation coefficient - 0.15% / °C).

[0279] Piezoelectric signal: Charge amplifier (Kistler 5114B) → 4 - th order Butterworth low - pass filter (cut - off frequency 500 Hz) → 16 - bit ADC (LTC1859).

[0280] Capacitive signal: Differential measurement circuit (excitation frequency 1 MHz, amplitude 5 V) → Phase - sensitive detector (AD630) → Digital lock - in amplifier (implemented by FPGA).

[0281] Data fusion algorithm:

[0282] Weighted fusion formula:

[0283] P fusion = 0.4P cap + 0.3Ppiezo +0.3P resistive ;

[0284] ·Basis for weight allocation:·Top capacitive sensor: Prioritize reflecting static pressure (weight 40%)·Middle piezoelectric sensor: Capture dynamic fluctuations (weight 30%)·Bottom piezoresistive sensor: Resist steam condensation interference (weight 30%)·Dynamic weight adjustment (abnormal conditions):

[0285]

[0286] When the standard deviation σi of a certain sensor exceeds the threshold value (such as ±5 kPa), its weight is automatically reduced to a minimum of 10%.

[0287] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A rice steaming vehicle energy-saving control method based on pressure feedback, characterized in that: The following steps are involved: Piezoresistive, piezoelectric and capacitive pressure sensors are arranged at the top, middle and bottom of the box 30cm from the bottom, a 24V DC power supply operation panel is set on the front side, and the internal electric heating tube is connected to the controller and integrated with the liquid level sensor; The heating mode is selected according to the interval distribution of the pressure value relative to the preset upper and lower limit thresholds, and the maximum steam pressure threshold is set to 0.1MPa. When overpressure occurs, the top air pressure valve is triggered to release pressure, including overpressure cut-off mode, low-pressure full-power mode and transition zone proportional adjustment mode, in which the upper and lower limit thresholds are dynamically adjusted according to the type of rice; When the steam pressure is ≥0.1MPa, all electric heating tubes are turned off. When the pressure is ≤0.09MPa, a single 4kW electric heating tube is started, and the cumulative heating time is ≥18 minutes within 30 minutes. In the transition zone mode, the PWM duty ratios of the basic group and enhanced group electric heating tubes are dynamically allocated based on the pressure change rate calculated in a 1-second sliding window; Synchronously collect the 32kHz sampling voiceprint signal of the dual microphone array and extract the energy proportion of the 500-1500Hz frequency band for leakage detection; Control 12 groups of circumferentially arranged nickel-titanium alloy actuators, with a diameter of 1.2-1.5 mm, a phase transition temperature of 105-110°C, and a deformation of 0-0.3 mm; The rice type was determined by classifying 60x microscopic images using the ResNet-18 network; Predict pressure decay trends based on LSTM networks and trigger level 3 safety responses; The TD3 reinforcement learning algorithm is used to optimize the control parameters, and the reward function weight ratio is energy efficiency 5: quality 3: loss 2.

2. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The sensor signal processing includes: The output of the piezoresistive sensor is temperature compensated with a compensation coefficient of -0.15% / ℃; The piezoelectric sensor signal is processed by a charge amplifier and a 4th-order Butterworth filter; The liquid level sensor has a range of 0-30cm and an accuracy of ±1mm. When the liquid level is ≤8cm, the solenoid valve is opened to replenish water, and when it is ≥25cm, it is closed. When the liquid level is <5cm, the power supply of the electric heating tube is forcibly cut off. The capacitive sensor uses a differential measurement circuit with an excitation frequency of 1MHz; The weighted fusion formula is: P fusion =0.4P 压阻 +0.3P 压电 +0.3P 电容 。 3. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The dynamically allocated electric heating tube group comprises: The electric heating tubes are divided into a basic group of 3kW×2, an enhanced group of 6kW×1 and an emergency group of 4kW×1; When the absolute value of the pressure change rate is greater than 5 kPa / s, the enhancement group is started first and works at 80% duty cycle PWM; In the range of 0.09-0.1MPa, the emergency group works in pulse mode.

4. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The voiceprint feature correction includes: extracting the energy proportion of the 500-1500Hz frequency band in the steam flow noise, and when the proportion is lower than a preset value, it is determined that steam leakage has occurred, triggering heating power compensation and sealing structure enhancement control.

5. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The spectrum feature correction includes: The voiceprint signal with a sampling rate of 32kHz is collected through a dual microphone array; Extract the energy percentage value of the 500-1500Hz frequency band, and determine it as a steam leak when the value is lower than 45%; Trigger the heating power compensation strategy: increase the current power output by 15%-25%. The compensation time is positively correlated with the degree of leakage and is no longer than 120 seconds.

6. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The optimal pressure parameter combination determination includes: Collect 60x microscope images of the rice grain surface and extract local binary pattern features and gray-level co-occurrence matrix features; Output rice types through convolutional neural network classifier and associate preset pressure parameters: Japonica rice: upper threshold 118±2kPa, hysteresis width 12kPa Indica rice: upper threshold 125±3kPa, hysteresis width 8kPa Glutinous rice: upper threshold value 110±1kPa, hysteresis width 15kPa.

7. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The hierarchical security mechanism includes: Level 1 response: When the pressure decay rate exceeds 0.8 kPa / min, an audible and visual alarm is triggered and a fault log is recorded; Secondary response: When the pressure is greater than 0.1 MPa for 5 seconds, the pressure relief valve is triggered and the power supply of the electric heating tube is cut off; Level 3 response: When the local temperature is detected to be over 130°C, inert gas is released and an emergency shutdown signal is uploaded.

8. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The reinforcement learning algorithm includes: Define the state space as a four-dimensional vector and the action space as a three-dimensional parameter adjustment; The reward function is designed as a weighted combination of energy efficiency factor, cooking quality score and equipment loss coefficient, with a weight ratio of 5:3:2; The strategy update cycle is to perform batch learning every 20 cooking tasks.

9. The energy-saving control method for a rice steaming vehicle based on pressure feedback according to claim 1, characterized in that: The method for enabling the standby redundant channel comprises: Disconnect the signal link of the suspected faulty sensor and inject a standard test pressure signal into the data bus; If the system pressure reading fluctuates by more than ±3kPa within 10 seconds, the sensor will be locked and switched to the adjacent sensor weighting mode; An automatic zero calibration is performed at the first start-up each day, and a reference pressure of 80 kPa is maintained during the calibration.

10. The energy-saving control method for a rice steaming cart based on pressure feedback according to claim 1, characterized in that: The control parameters of the nickel-titanium alloy actuator include: Driving voltage 12VDC, pulse width modulation frequency 8kHz; Temperature feedback sampling period 100ms, PID control parameter K p =2.5,K i =0.02,K d =0.8; The operation panel power supply circuit is equipped with overcurrent protection, with a maximum carrying current of 5A and a ripple voltage of <50mV; The deformation self-detection program is executed after each cooking, and the detection accuracy reaches 0.005mm.

Images