Consistency control system and method for cereal bar physical and chemical sugar spraying based on multi-sensor fusion

Abstract

This application relates to the technical field of food processing, and in particular to a control system and method for consistent physical and chemical spraying of cereal bars based on multi-sensor fusion. It includes a queue module that collects coded pulses, discretizes the physical space of the conveyor belt into a virtual grid, and constructs a follow-up queue; a feedforward module that, when a feedforward command is detected, extracts the jitter pulse width and converts the system's fixed physical delay time into a delayed pulse equivalent; compares the jitter pulse width with the delayed pulse equivalent to intercept or allow the feedforward command, and records a status tag; a feedback module that acquires the visual spraying thickness; if a status tag exists and the visual spraying thickness deviates from a thickness threshold, the visual spraying thickness is removed from the sample pool; otherwise, it is added to the sample pool; and an execution module that superimposes the opening signal and pressure signal to execute the spraying operation. This application can construct a bidirectional decoupled architecture, solving the problem of vicious cross-oscillations caused by spatiotemporal asynchrony, thereby achieving consistent physical and chemical spraying.

Description

Technical Field

[0001] This application relates to the technical field of food processing, and in particular to a control system and method for consistent physicochemical spraying of grain bars based on multi-sensor fusion. Background Technology

[0002] In modern industrial food processing, especially in the pretreatment stage of mixed grain layers on continuous, high-speed production lines for products like cereal bars, the consistency of the material surface coating directly determines the quality and economic benefits of the final product. With the widespread adoption of intelligent manufacturing equipment technology, production lines are typically equipped with various vision and motion sensors to achieve high-precision closed-loop control through real-time acquisition and fusion of multi-source information. In this high-speed dynamic processing scenario, the high-speed operation of the conveyor belt and the spraying action of the fluid material require extremely high spatiotemporal synchronization. Any minute mechanical disturbance or control signal delay can be amplified throughout the entire production line, leading to a large number of defective products.

[0003] A search revealed that Chinese Patent Publication No. CN1429668A discloses a prior art device called a machine vision fruit grading system controlled by a shift register. This system mainly includes components such as an encoder, a conveyor, a camera, and a shift register group. Its working principle is that the encoder, installed on the conveyor, generates a synchronization signal, causing the data in the shift register group to correspond to the physical position of the material. When the material moves to a specific position on the conveyor belt, the system acquires an image, which is processed by a computer to obtain the grading result based on external features. Subsequently, the control signal in the shift register moves synchronously with the material's movement, ultimately triggering the mechanical grading mechanism when the material reaches the designated discharge port. This technology effectively utilizes the combination of position encoding and data queuing, achieving synchronous tracking of spatial position and automatic grading control in the visual inspection and delayed execution tasks of discrete materials.

[0004] However, applying the existing control logic, primarily used for open-loop tracking of static properties of discrete materials, to scenarios requiring continuous dynamic closed-loop control, such as high-speed mixed grain layer fluid spraying, exposes a core technical problem: a vicious cross-oscillation caused by the strong coupling between feedforward spatiotemporal phase inversion and feedback steady-state and transient characteristics. Specifically, in high-speed conveying scenarios, the conveyor belt inevitably experiences brief, high-frequency mechanical vibrations. Upon detecting these transient fluctuations, the front-end sensor immediately triggers feedforward compensation. However, limited by the mechanical response of the solenoid valve and the fixed physical time delay of fluid flight, by the time the compensation fluid actually reaches the surface of the mixed grain layer, the high-frequency vibrations of the conveyor belt have often already entered the reverse recovery phase. This misalignment in physical space and time causes the old compensation action to occur on the new phase, not only failing to smooth out the fluctuations but also actively creating substandard products with localized excessive spraying. Subsequently, these locally defective products reach the back-end visual inspection area. Due to the lack of historical tracing capabilities of the existing spatially lagging visual feedback system, it cannot distinguish whether the defect stems from transient flaws caused by historical mechanical vibrations or steady-state drift caused by changes in the fluid's own physicochemical properties. The vision system often misjudges it as a global steady-state drift and generates widespread negative feedback, incorrectly lowering the baseline coating volume for the entire production line. Ultimately, the entire control system inevitably falls into an endless vicious cycle of feedforward misalignment leading to defective products, feedback misjudgment altering the baseline, and consequently, a continuous stream of defective products, making it impossible to achieve high-precision coating consistency control. Summary of the Invention

[0005] In order to construct a bidirectional decoupled architecture, solve the vicious cross-oscillation caused by spatiotemporal asynchrony, and thus achieve physical and chemical spraying consistency, this application provides a grain bar physical and chemical spraying consistency control system and method based on multi-sensor fusion.

[0006] Firstly, the cereal bar physicochemical spraying consistency control system based on multi-sensor fusion provided in this application adopts the following technical solution: the cereal bar physicochemical spraying consistency control system based on multi-sensor fusion includes: The queue module collects the coded pulses of the mixed grain layer conveyor belt, discretizes the physical space of the conveyor belt into a virtual grid, and constructs a follow-up queue with the virtual grid as the primary key. When the feedforward module detects conveyor belt jitter and triggers a feedforward command, it extracts the jitter pulse width, converts the fixed physical delay time of the system into a delay pulse equivalent, compares the jitter pulse width with the delay pulse equivalent to intercept or allow the feedforward command, and records a status label in the corresponding virtual grid. The feedback module obtains the visual sugar spraying thickness of the mixed grain layer and reads the status label corresponding to the virtual grid. If the status label exists and the visual sugar spraying thickness reaches a preset deviation thickness threshold, the visual sugar spraying thickness is removed from the sample pool for updating the steady-state benchmark; otherwise, it is added to the sample pool to update the steady-state benchmark. The execution module superimposes the opening signal generated based on the feedforward command for release with the pressure signal generated based on the updated steady-state reference, and coordinates the control of the end nozzle of the system to perform sugar spraying operation.

[0007] Optionally, the queue module is configured to: Based on a preset unit pulse equivalent, the virtual grid is configured as a first-in-first-out sequence; Identify the physical regions of each segment of the mixed grain layer on the conveyor belt, and establish a lifecycle binding between the physical regions and the uniquely corresponding virtual mesh; The motion detection status, valve action command, and visual detection result are sequentially acquired along the running direction of the conveyor belt, and the motion detection status, valve action command, and visual detection result are attached to the follow-up queue using the identifier of the virtual grid as the addressing primary key.

[0008] Optionally, the feedforward module is configured to: extract the jitter pulse width and convert the delayed pulse equivalent. The sum of the computation time of the system, the mechanical action time of the end nozzle, and the fluid flight time is obtained as the fixed physical delay time of the system. Based on the current speed of the conveyor belt and the unit pulse equivalent, the fixed physical delay time is converted into the delay pulse equivalent; Identify pulse intervals that change continuously in the same direction during the motion detection state, and extract their span as the jitter pulse width.

[0009] Optionally, the feedforward module is configured as follows: The interception threshold is set based on the delayed pulse equivalent; If the jitter pulse width is less than the interception threshold, the feedforward command is forcibly intercepted, and a transient interference-free label is recorded in the corresponding virtual grid as the status label. If the jitter pulse width is greater than or equal to the interception threshold, the feedforward instruction is allowed to proceed, and a trend-compensated label is recorded in the corresponding virtual grid as the status label. The status label is attached to the follow-up queue and configured for the feedback module to read back, serving as a logical condition for determining whether to remove the corresponding visual spray candy thickness from the sample pool.

[0010] Optionally, the feedforward module is further configured to: Construct a hysteresis comparison interval around the interception threshold; When the jitter pulse width falls into the hysteresis comparison interval, the status label of the previous adjacent virtual grid is read from the follow-up queue; The interception or release of the feedforward command is maintained according to the read status tag, and the same status tag is recorded in the corresponding virtual mesh.

[0011] Optionally, the feedback module is configured as follows: Backtrack the follow-up queue and read the status label corresponding to the virtual mesh; If the status label is identified as the transient interference-free label or the trend-compensated label, and the visual sugar spraying thickness reaches the deviation thickness threshold, a mask removal operation is triggered. Based on the mask culling operation, the visual candy thickness is removed from the sample pool used to calculate the global steady-state moving average.

[0012] Optionally, the feedback module is configured to update the steady-state reference as follows: If it is identified that the corresponding virtual grid does not record the transient interference-free label and the trend-compensated label, and the visual spray candy thickness reaches the deviation thickness threshold, the visual spray candy thickness is incorporated into the sample pool; The steady-state moving average deviation is calculated based on the sample pool filtered by the mask rejection operation. The steady-state reference is updated based on the steady-state moving average deviation and the preset adjustment step size.

[0013] Optionally, the execution module is configured to superimpose the opening signal and the pressure signal as follows: The transient valve opening of the end nozzle is controlled according to the opening signal; Adjust the base pressure of the fluid pipeline connected to the end nozzle according to the pressure signal; The base pressure and the transient valve opening are physically superimposed at the end nozzle to decouple the feedforward command for release from the updated steady-state reference.

[0014] Secondly, the cereal bar physicochemical spraying consistency control method based on multi-sensor fusion provided in this application adopts the following technical solution: Cereal bar physicochemical spraying consistency control method based on multi-sensor fusion, the method comprising: Based on the collected coded pulses, the physical space of the conveyor belt is discretized into the virtual grid, and the follow-up queue is constructed using the virtual grid as the primary key; By comparing the extracted jitter pulse width with the converted delayed pulse equivalent, the feedforward command is either forcibly intercepted or allowed to proceed, and the status label is attached to the corresponding virtual grid. Backtrack the state label in the follow-up queue. When the state label exists and the visual spray candy thickness reaches the deviation thickness threshold, remove the visual spray candy thickness from the sample pool used to calculate the steady-state benchmark. Otherwise, merge it into the sample pool to update the steady-state benchmark. The opening signal is generated based on the feedforward command for release, and the pressure signal is generated based on the updated steady-state reference. The opening signal and the pressure signal are superimposed to coordinately control the end nozzle of the system to perform sugar spraying operation.

[0015] Optionally, the method further includes: Based on the delayed pulse equivalent, an interception threshold is set, and a hysteresis comparison interval is constructed around the interception threshold. When the jitter pulse width falls into the hysteresis comparison interval, the status label of the previous adjacent virtual grid is traced back, and the interception or release of the feedforward command is maintained according to the status label. Calculate the steady-state moving average deviation of the sample pool and compare the steady-state moving average deviation with a preset dead zone threshold; When the steady-state moving average deviation reaches the dead zone threshold, the steady-state reference is updated based on the steady-state moving average deviation.

[0016] In summary, this application includes the following beneficial technical effects: 1. By constructing a follow-up queue with a virtual grid as the primary key, the data stream is spatiotemporally bound to the physical entity of the mixed grain layer throughout its entire lifecycle. Invalid high-frequency feedforward commands are precisely intercepted based on the comparison between the jitter pulse width and the equivalent of the delayed pulse. At the same time, an abnormal data caused by transient disturbances is removed from the sample pool using a status tag backtracking mechanism. This constructs a control architecture that decouples feedforward and feedback, fundamentally solving the core technical problem of vicious cross-oscillation caused by spatiotemporal asynchrony. Ultimately, it achieves highly consistent control of physical and chemical spraying of sugar before the subsequent grain bar forming.

[0017] 2. By discretizing the physical space of the conveyor belt into a virtual grid and establishing a follow-up queue, the motion detection status, valve action commands, and visual inspection results are all mounted and addressed using the virtual grid as the unique primary key. This achieves precise binding and seamless traceability of production process data with the physical location of materials, laying a solid spatiotemporal synchronization foundation for high-precision closed-loop control in high-speed dynamic processing scenarios.

[0018] 3. This scheme adopts a decoupled execution strategy that physically superimposes the feedforward opening signal and the feedback pressure signal, so that the high-frequency transient compensation and low-frequency steady-state adjustment are generated independently and superimposed only at the end nozzle. At the same time, the mask rejection operation is used to ensure that the steady-state reference is updated only based on the real drift data, effectively cutting off the interference link of transient disturbances to the global reference, and significantly enhancing the long-term stability and anti-interference capability of the control system under complex working conditions. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall hardware structure layout of the system in this application; Figure 2 This is a block diagram showing the module architecture and data flow of the system in this application; Figure 3 This is a schematic diagram of the mapping and data mounting principle of the follow-up queue based on the virtual grid in this application; Figure 4 This is a schematic diagram of the two-level control structure of the execution module and fluid pipeline in this application; Figure 5 This is a waveform diagram comparing the visual spraying thickness and control effect of the prior art and the control system of this application. Detailed Implementation

[0020] The following combination Figures 1-5 This application will be described in further detail.

[0021] This application discloses a grain bar physicochemical spraying consistency control system based on multi-sensor fusion, which is applied to the surface spraying process of the mixed grain layer in a continuous high-speed production line for grain bars, and provides targeted solutions to the core technical defects existing in the prior art.

[0022] In existing technologies, the solution disclosed in CN1429668A is representative. It uses encoder synchronous tracking and shift register queue control to bind the visual inspection results of discrete materials with their physical positions, thus completing open-loop grading. However, when this type of solution is applied to the continuous dynamic closed-loop control scenario of high-speed mixed grain layer fluid spraying, it cannot solve the vicious cross-oscillation problem caused by the strong coupling of feedforward spatiotemporal phase inversion and feedback steady-state and transient characteristics. Specifically, when the high-frequency mechanical vibration of the conveyor belt triggers feedforward compensation, due to the fixed physical delay of the system, by the time the compensation fluid reaches the material surface, the conveyor belt vibration has already entered the reverse recovery period, causing feedforward phase inversion and actively generating local spraying defects. The back-end visual inspection system cannot distinguish whether the defect is a transient defect caused by mechanical vibration or a steady-state drift caused by changes in fluid physicochemical properties. It incorrectly judges the transient defect as a steady-state drift and adjusts the global spraying benchmark, ultimately causing the system to fall into a vicious oscillation closed loop of feedforward misalignment causing defects, feedback misjudgment changing the benchmark, and continuous global defect generation, making it impossible to achieve high-precision spraying consistency control.

[0023] This solution constructs a full-process spatiotemporal synchronization architecture based on physical pulse drive to achieve bidirectional decoupling of feedforward and feedback, thereby cutting off the transmission path of malignant oscillations at the source and achieving stable control of the consistency of sugar spraying in the mixed grain layer. Figure 1 The data flow and signaling interaction between the queue module, feedforward module, feedback module, and execution module are illustrated in a block diagram. It clearly demonstrates the core architecture of this application's "bidirectional decoupling" and highlights the closed-loop backtracking role of status tags between modules. The following steps provide a detailed explanation: The S1 system collects the coded pulses from the mixed grain layer conveyor belt, discretizes the physical space of the conveyor belt into a virtual grid, and constructs a follow-up queue with the virtual grid as the primary key.

[0024] The system has a high-precision incremental rotary encoder coaxially mounted on the main drive shaft of the mixed grain conveyor belt. As the system's sole synchronous clock source, the encoder replaces the absolute time axis, which is prone to asynchronous time differences in traditional industrial control, and achieves spatiotemporal synchronization of all sensor signals, control actions, and the physical position of the conveyor belt. This avoids the control deviation problem caused by the misalignment of multiple data sources and physical entities in existing technologies from the infrastructure level.

[0025] Figure 2 The demonstration visually illustrates the physical relative positions of the main drive shaft rotary encoder, the area array visual positioning sensor at the feeding positioning station, the laser Doppler velocimetry sensor at the motion detection station, the end nozzle at the spraying station, and the linear array visual inspection sensor at the visual inspection station on the conveyor belt. It intuitively proves the hardware foundation of multi-sensor fusion, providing a realistic and feasible physical basis for calculating the "discrete physical space into a virtual mesh" and "fixed physical delay time" in the system.

[0026] In this embodiment, the encoder's pulse count per revolution is set to 2000 PPR, and the diameter of the main drive roller of the conveyor belt is set to 10cm. Each revolution of the main drive roller covers a straight-line distance of 31.416cm, which corresponds one-to-one with the 2000 complete pulses output by the encoder per revolution. Based on this correspondence, the system determines a preset unit pulse equivalent of 0.015708cm / pulse. This unit pulse equivalent setting matches the physical dimensions of the mixed grain layer and the positioning accuracy requirements of the production line. The typical length of a single segment of the mixed grain layer is 8cm to 15cm. This unit pulse equivalent ensures that the complete physical area of ​​a single segment of the mixed grain layer corresponds to at least 500 continuous virtual grids, achieving precise positioning of the mixed grain layer spraying area and deviation-free tracking throughout the entire process.

[0027] The system, based on a preset unit pulse equivalent, discretizes the continuous physical space of the conveyor belt into continuous virtual grids according to the pulse sequence output by the encoder. All virtual grids are configured in a first-in, first-out (FIFO) sequence according to the conveyor belt's travel direction, constructing a follow-up queue. Each time the encoder outputs a pulse, a new virtual grid is generated at the tail of the follow-up queue and completes its enqueue operation, while the virtual grid at the head of the queue synchronously completes its dequeue operation. The enqueue and dequeue actions are completely synchronized with the encoder pulse output. The total length of the follow-up queue is determined based on the physical distance from the motion detection station at the front end of the conveyor belt to the visual inspection station at the end. In this embodiment, the conveyor belt is sequentially equipped with a feeding positioning station, a motion detection station, a spraying station, and a visual inspection station along the travel direction. The physical distance between the motion detection station and the visual inspection station is 6 meters, corresponding to 38,200 virtual grids. This length setting ensures that the virtual grids bound to the mixed grain layer remain in the follow-up queue throughout the entire process from entering the motion detection station to completing the visual inspection, achieving full-process traceability of the production status.

[0028] The system uses an area array visual positioning sensor at the feeding and positioning station at the feed end to identify the physical regions of each segment of the mixed grain layer on the conveyor belt. The area array visual positioning sensor is installed directly above the feeding and positioning station, and its sampling frequency is perfectly synchronized with the encoder's pulse output frequency, ensuring that the identification result of the mixed grain layer's physical region perfectly matches the generation sequence of the virtual mesh. The system identifies the leading and trailing edges of the mixed grain layer's physical region, using the virtual mesh corresponding to the leading edge as the starting mesh and the virtual mesh corresponding to the trailing edge as the ending mesh, establishing a lifecycle binding between the complete physical region of a single segment of the mixed grain layer and the continuous, unique sequence of virtual meshes from the starting mesh to the ending mesh. For any offset or skew of the mixed grain layer on the conveyor belt, the system uses the circumscribed rectangle of the mixed grain layer's physical region as the boundary to complete the binding with the corresponding virtual mesh, ensuring the integrity of the binding relationship. Throughout the entire process of the mixed grain layer moving with the conveyor belt, the bounded virtual mesh sequence advances synchronously in the follow-up queue, and the correspondence between the two remains unchanged throughout the entire lifecycle. After the mixed grain layer leaves the visual inspection station, the corresponding virtual grid sequence completes its lifecycle and moves out synchronously with the queue.

[0029] Figure 3 It demonstrates how to map the physical regions (leading and trailing edges) of a grain bar onto a sequence of virtual grids driven by encoder pulses, and shows the process of attaching motion states, valve commands, and detection results to the grid primary key, which concretizes the abstract "first-in, first-out queue" with the "full life cycle binding" of materials.

[0030] The system sequentially acquires the motion detection status of the mixed grain layer, valve action commands, and visual inspection results along the conveyor belt's running direction, with the data acquisition order completely consistent with the workstation layout order. The system uses the unique identifier of the virtual grid as the addressing primary key, attaching all motion detection status, valve action commands, and visual inspection results to the corresponding virtual grid in the follow-up queue. This data attachment method enables unified collection and addressing of production data corresponding to the same physical location throughout the entire process, providing a unified benchmark for full-process traceability of status information during production and addressing the core deficiency of existing discrete material tracking schemes that cannot adapt to continuous dynamic closed-loop control.

[0031] When the S2 system detects conveyor belt jitter and triggers a feedforward command, it extracts the jitter pulse width, converts the system's fixed physical delay time into a delayed pulse equivalent, compares the jitter pulse width with the delayed pulse equivalent to intercept or allow the feedforward command, and records a status label in the corresponding virtual grid.

[0032] The system pre-calibrates a fixed physical delay time, which is the sum of the system's computation time, the mechanical action time of the end nozzle, and the fluid's airborne flight time. The system's computation time is calibrated using the programmable logic controller's (PLC) cyclic scan cycle, representing the entire process time for the controller to complete a single feedforward logic operation. The calibrated value is 2ms, covering the complete cycle of the industrial PLC from signal acquisition and logic judgment to instruction output, perfectly matching the pulse output timing of the encoder in S1. In this embodiment, the PLC's cyclic scan cycle is no greater than 1ms, ensuring the computation time remains stable within the calibrated range. The end nozzle's mechanical action time is calibrated using a solenoid valve step response test, representing the entire process time from the nozzle's high-speed solenoid valve receiving the action command to the valve opening changing to the target degree. The calibrated value is 5ms, matching the rated response performance of commonly used high-speed solenoid valves in the food coating industry, covering the complete process from solenoid valve coil excitation to the valve core completing its stroke. The flight time of the fluid in the air was calibrated using high-speed cameras. It is the flight time from when the sugar solution is ejected from the nozzle valve to when it reaches the upper surface of the mixed grain layer. The calibration is based on a vertical distance of 10 cm from the nozzle outlet to the upper surface of the mixed grain layer and a rated initial velocity of 5 m / s for the sugar solution. The calibration value is 20 ms. In this embodiment, the applicable sugar solution has a solid content of 70%-80% and a kinematic viscosity of 100-200 mPa·s, which matches the conventional process requirements of the grain bar sugar spraying process. The total calibration value of the fixed physical delay time of the system is 27 ms.

[0033] The system converts the fixed physical delay time into a delayed pulse equivalent based on the current operating speed of the conveyor belt and the unit pulse equivalent determined in S1. The rated operating speed of the conveyor belt is 60 m / min, corresponding to 1 m / s. The system first calculates the straight-line distance traveled by the conveyor belt within the fixed physical delay time, which is the product of the fixed physical delay time and the current operating speed of the conveyor belt. The system then calculates the number of pulses contained within this distance, which is the delayed pulse equivalent. At the rated operating speed, the distance traveled by the conveyor belt in 27 ms is 0.027 m, corresponding to a delayed pulse equivalent of 172 pulses. The system updates the delayed pulse equivalent synchronously with the real-time operating speed of the conveyor belt, ensuring that the spatial equivalent corresponding to the system's physical delay always precisely matches the motion state of the conveyor belt, while maintaining complete consistency with the virtual grid spatial reference constructed in S1.

[0034] The system collects real-time data on the conveyor belt's operating speed fluctuations using a laser Doppler velocity sensor installed at the motion detection station described in S1. The laser Doppler velocity sensor is installed 1m upstream of the spraying station, with a fixed physical distance from the nozzle at the end of the spraying station. This distance corresponds to a fixed number of virtual grids, ensuring precise matching between the timing of feedforward commands and the arrival time of the mixed grain layer at the nozzle. The system presets a feedforward command trigger threshold; when the real-time operating speed of the conveyor belt deviates from the rated operating speed by ±2%, a feedforward command is triggered, thus filtering out invalid command triggers caused by minor speed fluctuations. The system synchronously attaches the collected operating speed fluctuation data to the motion detection status item of the corresponding virtual grid in the follow-up queue, using the unit pulse equivalent determined in S1 as the timing reference. The system identifies pulse intervals exhibiting continuous unidirectional changes during motion detection. These intervals refer to the pulse sequence intervals corresponding to the duration of continuous increases or decreases in the conveyor belt speed. The system defines a valid continuous unidirectional speed change interval as three or more consecutive pulses of unidirectional speed change, and extracts the pulse span of this interval as the jitter pulse width. The system uses this to distinguish between high-frequency reciprocating jitter and long-term speed fluctuations in the conveyor belt, providing a basis for judging the interception and release of feedforward commands, and fundamentally avoiding the phase misalignment problem caused by indiscriminate triggering of feedforward compensation in existing technologies.

[0035] The system sets an interception threshold based on the delayed pulse equivalent, which is set to twice the delayed pulse equivalent. The threshold setting is based on the matching relationship between the conveyor belt's jitter motion characteristics and the system's fixed physical delay. When the jitter pulse width is less than twice the delayed pulse equivalent, the conveyor belt's jitter trend will enter a reverse recovery period before the compensated sugar solution reaches the surface of the mixed grain layer. At this time, performing feedforward compensation will result in a feedforward phase inversion problem, leading to overcompensation. In this embodiment, at the rated operating speed, the delayed pulse equivalent is 172 pulses, corresponding to an interception threshold of 344 pulses.

[0036] The system compares the extracted jitter pulse width with the set interception threshold in real time. If the jitter pulse width is less than the interception threshold, the system forcibly intercepts the feedforward command, does not send a feedforward compensation action signal to the end nozzle, and maintains the steady-state output of the nozzle. Simultaneously, the system records a transient interference-free tag as a status tag in the virtual grid corresponding to the jitter event in the follow-up queue constructed in S1. If the jitter pulse width is greater than or equal to the interception threshold, the system allows the feedforward command to be released and sends the corresponding feedforward compensation action signal to the end nozzle. The valve opening adjustment range of the feedforward compensation action signal is linearly related to the jitter pulse width. When the conveyor belt speed continuously increases, the valve opening increases synchronously; when the conveyor belt speed continuously decreases, the valve opening decreases synchronously. For every 10 pulses increase in the jitter pulse width, the valve opening is adjusted by 1%, with the upper limit of the adjustment range being ±20% of the full valve opening. Simultaneously, the system records a trend and compensation tag as a status tag in the virtual grid corresponding to the jitter event in the follow-up queue constructed in S1. All status tags are attached to the corresponding virtual grid of the follow queue, and advance synchronously with the entire life cycle of the virtual grid, always remaining bound to the physical location of the corresponding mixed grain layer.

[0037] The system constructs a hysteresis comparison interval around the interception threshold, with an upper limit of 105% and a lower limit of 95% of the interception threshold. This interval setting avoids logical oscillations caused by frequent switching between interception and release actions when the jitter pulse width fluctuates slightly around the interception threshold. The interval width is set to match the normal fluctuation amplitude of conveyor belt jitter, balancing logical stability and judgment accuracy. When the jitter pulse width rises from below the lower limit of the interval to within the hysteresis comparison interval, the system maintains the interception action for the feedforward command; when the jitter pulse width falls from above the upper limit of the interval to within the hysteresis comparison interval, the system maintains the release action for the feedforward command; when the jitter pulse width remains within the hysteresis comparison interval, the system reads the status label of the adjacent virtual grid in the direction of the conveyor belt's travel from the follow-up queue constructed in S1, maintains the interception or release action for the feedforward command according to the read status label, and records the same status label in the corresponding virtual grid.

[0038] The S3 system obtains the visual sugar spraying thickness of the mixed grain layer, reads the status label of the corresponding virtual grid, and if a status label exists and the visual sugar spraying thickness reaches the preset deviation thickness threshold, the visual sugar spraying thickness is removed from the sample pool for updating the steady-state benchmark; otherwise, it is added to the sample pool to update the steady-state benchmark.

[0039] The system uses a linear array visual inspection sensor installed at the visual inspection station S1, downstream of the spraying station, to collect data on the sugar spraying thickness of the mixed grain layer surface reaching the visual inspection area. The linear array visual inspection sensor employs laser triangulation to achieve high-precision detection of the sugar spraying thickness, with a detection accuracy of 0.01mm. Its sampling frequency is completely synchronized with the pulse output frequency of the encoder in S1, with a rated sampling frequency of no less than 10kHz, fully covering the encoder's pulse output frequency. This ensures that each virtual grid in the follow-up queue corresponds to a unique set of sugar spraying thickness detection data. The system synchronously attaches the collected sugar spraying thickness data to the follow-up queue constructed in S1, indexed by the unique identifier of the corresponding virtual grid. This achieves precise binding between the detection data and the physical position of the mixed grain layer, eliminating the judgment deviation caused by misalignment between detection data and material position in existing technologies. In this embodiment, the standard value for the sugar spraying thickness of the mixed grain layer is set to 0.5mm, matching the industry's standard pre-coating requirements for grain bar products.

[0040] The system has a preset thickness deviation threshold, which is set to ±15% of the standard value for spray coating thickness. This thickness deviation threshold is set to match the allowable thickness deviation requirements for subsequent molding of the mixed grain layer. The national standard allows a deviation of ±20% for the overall coating thickness of the grain bar product. Setting the thickness deviation threshold to ±15% allows for early identification of abnormal thickness data within the product's acceptable range, while avoiding misclassifying normal production fluctuations as abnormal data, thus ensuring the accuracy of subsequent steady-state baseline updates.

[0041] After acquiring the sugar spraying thickness detection data, the system uses the unique identifier of the virtual grid corresponding to the data set as the addressing primary key to backtrack the follow-up queue constructed in S1, read the status label recorded in step S2 within the corresponding virtual grid, and complete the identification of the cause of the abnormal sugar spraying thickness data. By tracing the source of the status label, the system distinguishes whether the thickness anomaly is caused by transient disturbances from the mechanical vibration of the conveyor belt or by systematic steady-state drift caused by changes in the physicochemical properties of the sugar solution itself, thus fundamentally avoiding the baseline misadjustment problem caused by the indiscriminate judgment of abnormal data by the feedback system in existing technologies.

[0042] The system performs a unified filtering logic on each set of acquired visual candy spray thickness data. If the system identifies a transient interference-free label or a trend-compensated label on the corresponding virtual mesh, and the visual candy spray thickness of the corresponding virtual mesh reaches a preset deviation thickness threshold, the system triggers a mask rejection operation. Based on the mask rejection operation, the system marks the set of visual candy spray thickness data as invalid data using Boolean flags and removes it from the sample pool used to calculate the global steady-state moving average. This avoids abnormal thickness data caused by mechanical transient disturbances or feedforward compensation actions from contaminating the steady-state benchmark calculation process and cuts off the interference link of feedforward actions on the feedback loop. Except for the above rejection cases, all other visual candy spray thickness data are incorporated into the sample pool, including data with state labels but thickness not reaching the deviation thickness threshold, data without state labels but thickness reaching the deviation thickness threshold, and data without state labels and thickness not reaching the deviation thickness threshold, ensuring that the sample pool can completely reflect the true steady-state operating state of the system.

[0043] The system employs a sliding window mode to set the sample pool, with the window length set to the thickness data of the past 60,000 valid virtual grids. This window length is matched to the production line's rated speed; at a rated operating speed of 60 m / min, the conveyor belt outputs approximately 6366 pulses per second. The 60,000 valid data points correspond to approximately 10 seconds of production time, simultaneously ensuring both the anti-interference stability of the steady-state baseline and the responsiveness to real changes in the physicochemical properties of the sugar solution. The sliding window slides synchronously with the encoder pulses; for each new valid virtual grid thickness data point, the window slides forward one virtual grid, maintaining a fixed total data volume within the window. Based on the sample pool filtered through a masking process, the system calculates the global steady-state moving average of all valid data within the pool, then calculates the difference between the global steady-state moving average and the standard value of the sugar spraying thickness to obtain the steady-state moving average deviation.

[0044] The system has a preset dead zone threshold, set at ±2% of the standard sugar spraying thickness. This dead zone threshold is designed to prevent system oscillations caused by frequent adjustments to the steady-state baseline due to minor normal fluctuations in sugar spraying thickness. The dead zone threshold is matched to the typical production fluctuation range of grain bar spraying, balancing system stability and control accuracy. The system compares the calculated steady-state moving average deviation with the preset dead zone threshold. When the steady-state moving average deviation reaches the dead zone threshold, the system updates the steady-state baseline based on the deviation and the preset adjustment step size. When the deviation does not reach the dead zone threshold, the system maintains the current steady-state baseline and does not perform any adjustment. The preset adjustment step size is positively correlated with the absolute value of the steady-state moving average deviation. The maximum value of the adjustment step size is set to 3% of the current steady-state baseline value to ensure smooth adjustment of the steady-state baseline, avoid spraying flow fluctuations caused by large adjustments, and maintain stable system operation.

[0045] The S4 system superimposes the opening signal generated based on the feedforward release command with the pressure signal generated based on the updated steady-state reference, and coordinates the end nozzle of the control system to perform sugar spraying operation.

[0046] The system generates an opening signal based on the feedforward command for permission to release in S2. This opening signal is a high-frequency transient control signal, directly transmitted to the high-speed solenoid valve at the end nozzle to control the transient valve opening. The high-speed solenoid valve is installed at the inlet of the nozzle, corresponding one-to-one with each nozzle, with a rated response frequency of no less than 10kHz, fully covering the encoder's pulse output frequency, thus meeting the response requirements for high-frequency transient opening adjustment. The generation of the opening signal is based on the timing of the corresponding virtual grid in the follow-up queue constructed in S1. The timing of the signal transmission is completely synchronized with the queue advancement of the virtual grid, ensuring that the valve opening adjustment action precisely matches the time when the mixed grain layer bound to the corresponding virtual grid arrives directly below the nozzle. The adjustment range of the opening signal is linearly related to the width of the jitter pulse extracted in S2. The larger the width of the jitter pulse, the greater the fluctuation range of the long-term speed of the conveyor belt. The adjustment range of the valve opening is synchronously adapted to respond to the fluctuation of the long-term speed of the conveyor belt, thereby achieving transient and accurate compensation of the spraying flow rate. This avoids the spraying deviation caused by the misalignment of the feedforward action and the material position, and avoids the core defect of feedforward phase inversion in the existing technology from the execution end.

[0047] The system generates a pressure signal based on the updated steady-state reference in S3. This pressure signal is a low-frequency steady-state control signal, transmitted to a pneumatic proportional valve at the front end of the fluid pipeline connecting to the terminal nozzle, regulating the base pressure of the fluid pipeline. The pneumatic proportional valve is installed on the main sugar solution pipeline 5m upstream of the spraying station, with a rated response frequency of 1Hz to 5Hz. This frequency range matches the update cycle of the steady-state reference in S3, with a minimum update cycle of 1s, suitable for low-frequency steady-state control requirements. The pressure signal is generated solely based on the steady-state moving average deviation calculated from the sample pool filtered by the mask rejection operation in S3. It only responds to systematic steady-state drift caused by changes in the physicochemical properties of the sugar solution and is unaffected by transient spraying deviations caused by conveyor belt mechanical vibration. The pressure signal has a linear relationship with the steady-state reference. For every 1% increase in the steady-state reference, the pipeline base pressure increases by 1% accordingly. The adjustment step size of the pressure signal strictly follows the preset adjustment step size in S3. The maximum adjustment amount in a single instance does not exceed 3% of the current value of the steady-state reference. This achieves smooth adjustment of the pipeline base pressure, ensures long-term stable control of the base sugar spraying volume, and avoids the global reference offset problem caused by feedback misjudgment in the existing technology.

[0048] The system's fluid piping employs a two-stage independent control structure. A pneumatic proportional valve at the front end of the piping regulates the base pressure of the sugar solution throughout the pipeline, responding to low-frequency steady-state control requirements. A high-speed solenoid valve at the end of each spraying station regulates the transient nozzle opening, responding to high-frequency transient control requirements. The two control stages are independent of each other, receiving pressure and opening signals separately, respectively. The signal transmission links are isolated to prevent coupling interference between the two types of control signals during transmission.

[0049] Figure 4 A two-stage fluid pipeline structure with an upstream pneumatic proportional valve and a terminal high-speed solenoid valve was drawn, which visually demonstrates how the low-frequency base pressure and the high-frequency transient opening signal are physically independent of each other and ultimately physically superimposed at the terminal nozzle.

[0050] The system physically superimposes the pipeline's base pressure and the nozzle's transient valve opening at the end nozzle. The final injection flow rate of the sugar solution is jointly determined by the pipeline's base pressure and the valve opening. The base pressure determines the base flow rate range for the sugar solution injection, while the valve opening determines the dynamic adjustment range of the instantaneous flow rate. High-frequency opening signals and low-frequency pressure signals are generated and transmitted independently, and are physically superimposed only at the sugar injection action of the end nozzle. This decouples the feedforward command for response release from the updated steady-state reference, cuts off the cross-oscillation link between the feedforward transient action and the feedback steady-state adjustment, consolidates the system's control stability from the execution end, and completely breaks the vicious oscillation loop formed by mutual interference between feedforward and feedback in existing technologies.

[0051] The system synchronously mounts the valve action command and actual execution result for each sugar spraying action to the follow-up queue constructed by S1, using the unique identifier of the corresponding virtual grid as the addressing primary key. The full lifecycle recording of valve action commands and execution results enables full-process traceability of the sugar spraying action, further ensuring the traceability of the spraying process and the closed-loop verification of the control logic, while providing complete data support for subsequent production process optimization and anomaly troubleshooting.

[0052] Figure 5 A comparison was made between the existing technology, which, after encountering mechanical vibration disturbances, falls into a continuously amplified "malicious cross-oscillation" waveform, and the present application's system, which, after initiating feedforward interception and mask removal, quickly maintains the thickness within the dead zone threshold in a stable waveform. This directly demonstrates that the present invention overcomes the serious defects of the prior art and achieves "unexpectedly superior technical effects."

[0053] The implementation principle of the grain bar physical-chemical spraying consistency control system based on multi-sensor fusion in this application is as follows: This application realizes the full life cycle spatiotemporal binding of data stream and physical entity of mixed grain layer by constructing a follow-up queue with virtual grid driven by coded pulse as the main key. This fundamentally solves the control deviation problem caused by misalignment of multi-source data and material position, and provides a unified synchronization benchmark for the whole process control. On this basis, by converting the fixed physical delay of the system into the equivalent of the delayed pulse and comparing it with the extracted jitter pulse width, invalid high-frequency feedforward commands are accurately intercepted. This avoids the technical defect of actively creating local defects due to feedforward spatiotemporal phase inversion from the source, and records the status label for abnormal data tracing. Subsequently, by tracing back the status label and triggering the mask rejection operation, The system precisely removes abnormal thickness data caused by transient disturbances such as mechanical vibration from the sample pool used to calculate the global steady-state benchmark, and updates the benchmark only based on the true steady-state drift data. This completely cuts off the vicious feedback misadjustment link where the feedback system misjudges transient defects as steady-state drift and thus erroneously lowers the global benchmark. Finally, the execution module physically superimposes and decouples the high-frequency opening signal generated based on the effective feedforward command and the low-frequency pressure signal generated based on the pure steady-state benchmark. This allows the feedforward transient compensation and feedback steady-state adjustment to work synergistically at the end nozzle rather than interfere with each other. This constructs a complete architecture of bidirectional decoupling between feedforward and feedback, completely solving the vicious cross-oscillation problem caused by spatiotemporal asynchrony mentioned in the background technology, and realizing high-precision consistency control of the physicochemical spraying of mixed grain layers.

[0054] This application also discloses a method for controlling the consistency of physicochemical spraying of sugar on cereal bars based on multi-sensor fusion, which is applied to the aforementioned physicochemical spraying of sugar on cereal bars control system based on multi-sensor fusion. The specific implementation steps of the method are as follows.

[0055] The S1 system collects the coded pulses from the mixed grain layer conveyor belt, discretizes the physical space of the conveyor belt into a virtual grid, and constructs a follow-up queue with the virtual grid as the primary key.

[0056] The system coaxially mounts a high-precision incremental rotary encoder on the main drive shaft of the mixed grain layer conveyor belt. The encoder's output pulses serve as the system's sole synchronous clock source, replacing the traditional absolute time axis which is prone to asynchronous time differences. This achieves spatiotemporal synchronization of all sensor signals, control actions, and the physical position of the conveyor belt. In this embodiment, the encoder's pulse count per revolution is set to 2000 PPR. The main drive roller of the conveyor belt has a diameter of 10cm. Each revolution of the main drive roller covers a linear distance of 31.416cm for the conveyor belt, corresponding one-to-one with the 2000 complete pulses output by the encoder per revolution. Based on this, the system determines a preset unit pulse equivalent of 0.015708cm / pulse. This unit pulse equivalent matches the typical mixed grain layer product size of 8cm to 15cm and the positioning accuracy requirements of the production line, ensuring that a complete physical area of ​​a single mixed grain layer corresponds to at least 500 continuous virtual grids, achieving precise positioning and full-process tracking of the spraying area.

[0057] Based on a preset unit pulse equivalent, the system discretizes the continuous physical space of the conveyor belt into continuous virtual grids according to the encoder pulse sequence. All virtual grids are configured as a first-in-first-out (FIFO) follow-up queue according to the conveyor belt's travel direction. Each time the encoder outputs a pulse, a new virtual grid is generated at the tail of the follow-up queue and added to the queue, while the virtual grid at the head of the queue is dequeued synchronously. The queue movement is completely synchronized with the encoder pulse output. The total length of the follow-up queue is determined based on the physical distance between the motion detection station at the front end of the conveyor belt and the visual inspection station at the end. In this embodiment, the physical distance is 6m, corresponding to 38,200 virtual grids. This ensures that the bound virtual grids remain in the follow-up queue throughout the entire process of the mixed grain layer from entering the motion detection area to completing visual inspection, achieving full-process traceability of the production status.

[0058] The system uses an area array visual positioning sensor at the feeding positioning station to identify the physical area of ​​each segment of mixed grain layer on the conveyor belt. The area array visual positioning sensor is installed directly above the feeding positioning station, and its sampling frequency is perfectly synchronized with the encoder pulse output frequency, ensuring that the identification results match the timing of virtual mesh generation. The system identifies the leading and trailing edges of the physical area of ​​the mixed grain layer, using the virtual mesh corresponding to the leading edge as the starting mesh and the virtual mesh corresponding to the trailing edge as the ending mesh. A lifecycle-long binding is established between the complete physical area of ​​a single segment of mixed grain layer and the continuous, unique virtual mesh sequence from the starting to the ending mesh. For cases of mixed grain layer offset or skew, the system uses the bounding rectangle of its physical area as the boundary to complete the binding, ensuring the integrity of the binding relationship. As the mixed grain layer moves with the conveyor belt, the bound virtual mesh sequence advances synchronously in the follow-up queue, and the correspondence between the two remains unchanged throughout the lifecycle. After the mixed grain layer leaves the visual inspection station, the corresponding virtual mesh sequence completes its lifecycle and moves out synchronously with the queue.

[0059] The system sequentially acquires the motion detection status of the mixed grain layer, valve action commands, and visual inspection results along the conveyor belt's running direction, with the data acquisition order completely consistent with the workstation layout order. Using a unique virtual grid identifier as the addressing primary key, the system attaches all the above data to the corresponding virtual grid of the follow-up queue, achieving unified collection and addressing of production data for the entire process at the same physical location, providing a unified benchmark for full-process traceability of production status.

[0060] When the S2 system detects conveyor belt jitter and triggers a feedforward command, it extracts the jitter pulse width, converts the system's fixed physical delay time into a delayed pulse equivalent, compares the jitter pulse width with the delayed pulse equivalent to intercept or allow the feedforward command, and records a status label in the corresponding virtual grid.

[0061] The system pre-calibrates a fixed physical delay time, which is the sum of the system's computation time, the end nozzle's mechanical action time, and the fluid's airborne flight time. The system computation time was calibrated using the cyclic scanning cycle of the programmable logic controller (PLC), representing the entire process time for the controller to complete a single feedforward logic operation. The calibrated value was 2ms, covering the complete cycle of the industrial PLC from signal acquisition and logic judgment to instruction output, perfectly matching the encoder pulse output timing. The mechanical action time of the end nozzle was calibrated using the solenoid valve step response test, representing the entire process time from receiving the action command to the valve opening changing to the target degree. The calibrated value was 5ms, matching the rated response performance of high-speed solenoid valves commonly used in the food coating industry. The fluid flight time was calibrated using high-speed imaging, representing the flight time of the sugar solution from the nozzle valve outlet to the surface of the mixed grain layer. Based on the vertical distance of 10cm from the nozzle outlet to the surface of the mixed grain layer and the rated initial injection velocity of 5m / s for the sugar solution, the calibrated value was 20ms, suitable for the sugar solution process requirements of 70%-80% solids content and 100-200mPa·s in this embodiment. The total calibrated value of the system's fixed physical delay time was 27ms.

[0062] The system converts a fixed physical delay time into a delayed pulse equivalent based on the current operating speed of the conveyor belt and the unit pulse equivalent determined by S1. The rated operating speed of the conveyor belt is 60 m / min, corresponding to 1 m / s. The system first calculates the straight-line distance traveled by the conveyor belt within the fixed physical delay time; this distance is the product of the fixed physical delay time and the current operating speed of the conveyor belt. Then, it calculates the number of pulses contained within this distance to obtain the delayed pulse equivalent. At the rated operating speed, the conveyor belt travels 0.027 m within 27 ms, corresponding to a delayed pulse equivalent of 172 pulses. The system synchronously updates the delayed pulse equivalent with the real-time operating speed of the conveyor belt, ensuring that the spatial equivalent corresponding to the system's physical delay always precisely matches the conveyor belt's motion state and is completely consistent with the virtual mesh spatial reference constructed by S1.

[0063] The system uses a laser Doppler velocimeter installed at the motion detection station (S1) to collect real-time data on conveyor belt speed fluctuations. The laser Doppler velocimeter is installed 1m upstream of the spraying station, with a fixed physical distance from the nozzle at the end of the spraying station, corresponding to a fixed number of virtual grids. This ensures precise matching between the timing of feedforward commands and the arrival time of the mixed grain layer at the nozzles. The system presets a feedforward command trigger threshold. When the real-time conveyor belt speed deviates from the rated speed by ±2%, a feedforward command is triggered, filtering out invalid command triggers caused by minor speed fluctuations. The system synchronously attaches the collected speed fluctuation data to the motion detection status item of the corresponding virtual grid in the follow-up queue, using the unit pulse equivalent determined in S1 as the timing reference. The system identifies pulse intervals that change continuously in the same direction during motion detection. These intervals refer to the pulse sequence intervals corresponding to the duration of continuous increase or decrease in the conveyor belt speed. The system sets three or more consecutive pulses of speed change in the same direction as valid continuous speed change intervals. The number of pulses in this interval is extracted as the jitter pulse width to distinguish between high-frequency reciprocating jitter of the conveyor belt and long-term speed fluctuations, providing a basis for the interception and release of feedforward commands.

[0064] The system sets an interception threshold based on the delayed pulse equivalent, which is set to twice the delayed pulse equivalent. This threshold setting is based on the matching relationship between the conveyor belt jitter motion characteristics and the system's fixed physical delay. When the jitter pulse width is less than twice the delayed pulse equivalent, the conveyor belt jitter trend will enter a reverse recovery period before the compensation sugar solution reaches the surface of the grain bars. At this time, performing feedforward compensation will result in phase inversion, causing reverse overcompensation. In this embodiment, the interception threshold corresponds to 344 pulses at the rated operating speed.

[0065] The system compares the extracted jitter pulse width with the set interception threshold in real time. If the jitter pulse width is less than the interception threshold, the system forcibly intercepts the feedforward command and does not send a feedforward compensation action signal to the end nozzle, maintaining the steady-state output of the nozzle. At the same time, in the virtual grid corresponding to the jitter event in the follow-up queue constructed by S1, a transient interference-free label is recorded as a status label. If the jitter pulse width is greater than or equal to the interception threshold, the system allows the feedforward command to be released and sends the corresponding feedforward compensation action signal to the end nozzle. The valve opening adjustment range of the feedforward compensation action signal is linearly corresponding to the jitter pulse width. When the conveyor belt running speed continuously increases, the valve opening increases synchronously; when it continuously decreases, the valve opening decreases synchronously. For every 10 pulses increase in the jitter pulse width, the valve opening is adjusted by 1%, with the upper limit of the adjustment range being ±20% of the full valve opening. At the same time, the system records the trend and compensation label as a status label in the virtual grid corresponding to the jitter event. All status tags are attached to the corresponding virtual grid in the follow queue, and advance synchronously with the entire life cycle of the virtual grid, always remaining bound to the physical location of the corresponding grain bar.

[0066] The system constructs a hysteresis comparison interval around the interception threshold. The upper limit of the hysteresis comparison interval is 105% of the interception threshold, and the lower limit is 95% of the interception threshold. This avoids logical oscillations caused by frequent switching between interception and release actions when the jitter pulse width fluctuates slightly around the interception threshold. When the jitter pulse width rises from below the lower limit of the interval to within the hysteresis comparison interval, the system maintains the interception action for the feedforward command; when the jitter pulse width falls from above the upper limit of the interval to within the hysteresis comparison interval, the system maintains the release action for the feedforward command; when the jitter pulse width remains within the hysteresis comparison interval, the system reads the status tag of the adjacent virtual grid in the direction of the conveyor belt's movement from the follow-up queue constructed in S1, maintains the interception or release action for the feedforward command according to the read status tag, and records the same status tag in the current corresponding virtual grid.

[0067] The S3 system obtains the visual sugar spraying thickness of the cereal bar, reads the status label of the corresponding virtual grid, and if a status label exists and the visual sugar spraying thickness reaches the preset deviation thickness threshold, the visual sugar spraying thickness is removed from the sample pool for updating the steady-state benchmark; otherwise, it is added to the sample pool to update the steady-state benchmark.

[0068] The system uses a linear array visual inspection sensor installed at the visual inspection station S1, downstream of the spraying station, to collect data on the sugar spraying thickness on the surface of the cereal bars arriving at the visual inspection area. The linear array visual inspection sensor employs laser triangulation to achieve high-precision detection with an accuracy of 0.01 mm. Its sampling frequency is completely synchronized with the pulse output frequency of the encoder in S1, with a rated sampling frequency of no less than 10 kHz, fully covering the encoder's pulse output frequency. This ensures that each virtual grid in the follow-up queue corresponds to a unique set of sugar spraying thickness detection data. The system synchronously attaches the collected sugar spraying thickness data to the follow-up queue constructed in S1, indexed by the unique identifier of the corresponding virtual grid, achieving precise binding between the detection data and the physical position of the cereal bars. In this embodiment, the standard value for the sugar spraying thickness of the cereal bars is set to 0.5 mm, matching the industry's standard spraying requirements for cereal bar products.

[0069] The system has a preset thickness deviation threshold, which is set to ±15% of the standard value of the sprayed sugar thickness. This threshold matches the national standard allowable deviation of ±20% for the sprayed sugar thickness of cereal bar products. It can identify abnormal thickness data in advance within the product's acceptable range, while avoiding misjudging normal production fluctuations as abnormal data, thus ensuring the accuracy of subsequent steady-state baseline updates.

[0070] After the system obtains the sugar spraying thickness detection data, it uses the unique identifier of the virtual grid corresponding to the data set as the addressing primary key, backtracks the follow-up queue constructed in S1, reads the status label recorded by step S2 in the corresponding virtual grid, completes the identification of the cause of the abnormal sugar spraying thickness data, and distinguishes whether the thickness abnormality is caused by the transient disturbance of the conveyor belt mechanical vibration or the systematic steady-state drift caused by the change of the physicochemical properties of the sugar liquid itself.

[0071] The system performs a unified filtering logic on each set of acquired visual candy spraying thickness data. If a transient interference-free label or a trend-compensated label is identified for the corresponding virtual mesh, and the visual candy spraying thickness of the virtual mesh reaches a preset deviation thickness threshold, the system triggers a mask rejection operation. This operation marks the set of visual candy spraying thickness data as invalid data using Boolean tags and removes it from the sample pool used to calculate the global steady-state moving average. This prevents abnormal thickness data caused by mechanical transient disturbances or feedforward compensation actions from contaminating the steady-state benchmark calculation process and cuts off the interference link between feedforward actions and the feedback loop. Except for the above rejection cases, all other visual candy spraying thickness data are incorporated into the sample pool, including data with state labels but whose thickness has not reached the deviation thickness threshold, data without state labels but whose thickness has reached the deviation thickness threshold, and data without state labels but whose thickness has not reached the deviation thickness threshold. This ensures that the sample pool can completely reflect the true steady-state operating state of the system.

[0072] The system employs a sliding window mode to set the sample pool, with the sliding window length set to the thickness data of the past 60,000 valid virtual grids. This window length matches the rated operating speed of the production line. At a rated operating speed of 60 m / min, the conveyor belt outputs approximately 6366 pulses per second. The 60,000 valid data sets correspond to a production line duration of approximately 10 seconds, simultaneously ensuring the anti-interference stability of the steady-state baseline and the response speed to real changes in the physicochemical properties of the sugar solution. The sliding window slides synchronously with the encoder pulses. For each newly added thickness data of a valid virtual grid, the window slides forward one virtual grid, always maintaining a fixed total amount of data within the window. Based on the sample pool filtered by the mask rejection operation, the system calculates the global steady-state moving average of all valid data within the sample pool, and then calculates the difference between the global steady-state moving average and the standard value of the sugar spraying thickness to obtain the steady-state moving average deviation.

[0073] The system has a preset dead zone threshold, set at ±2% of the standard value of the spray coating thickness. This prevents system oscillations caused by frequent adjustments to the steady-state reference due to minor normal fluctuations in the spray coating thickness. The system compares the calculated steady-state moving average deviation with the preset dead zone threshold. When the steady-state moving average deviation reaches the dead zone threshold, the system updates the steady-state reference based on the deviation and the preset adjustment step size. When the deviation does not reach the dead zone threshold, the system maintains the current steady-state reference and does not perform any adjustment. The preset adjustment step size is positively correlated with the absolute value of the steady-state moving average deviation. The maximum adjustment step size is set to 3% of the current steady-state reference value to ensure smooth adjustment of the steady-state reference and avoid spray flow fluctuations caused by large adjustments.

[0074] The S4 system superimposes the opening signal generated based on the feedforward release command with the pressure signal generated based on the updated steady-state reference, and coordinates the end nozzle of the control system to perform sugar spraying operation.

[0075] The system generates an opening signal based on the feedforward command for permission to release in S2. This opening signal is a high-frequency transient control signal, directly transmitted to the high-speed solenoid valve at the end nozzle to control the transient valve opening of the nozzle. The high-speed solenoid valve is installed at the inlet of the nozzle, corresponding one-to-one with each nozzle, with a rated response frequency of no less than 10kHz, fully covering the encoder pulse output frequency to meet the response requirements of high-frequency transient opening adjustment. The generation of the opening signal is based on the timing of the corresponding virtual grid in the follow-up queue constructed in S1. The timing of the signal release is completely synchronized with the queue advancement of the virtual grid, ensuring that the valve opening adjustment action is precisely matched with the time when the grain bar bound to the corresponding virtual grid arrives directly below the nozzle. The adjustment amplitude of the opening signal is linearly correlated with the width of the jitter pulse extracted in S2. The larger the jitter pulse width, the greater the long-term speed fluctuation amplitude of the conveyor belt. The valve opening adjustment amplitude is synchronously adapted to respond to the long-term speed fluctuation of the conveyor belt, thereby achieving transient and precise compensation for the spray flow rate.

[0076] The system generates a pressure signal based on the updated steady-state reference in S3. This pressure signal is a low-frequency steady-state control signal, transmitted to a pneumatic proportional valve at the front end of the fluid pipeline connected to the terminal nozzle, adjusting the base pressure of the fluid pipeline. The pneumatic proportional valve is installed on the main sugar solution pipeline 5m upstream of the spraying station, with a rated response frequency of 1Hz to 5Hz, matching the update cycle of the steady-state reference in S3 and adapting to low-frequency steady-state control requirements. The generation of the pressure signal is solely based on the steady-state moving average deviation calculated from the sample pool filtered by the mask rejection operation in S3. It only responds to the systematic steady-state drift caused by changes in the physicochemical properties of the sugar solution and is unaffected by transient spraying deviations caused by conveyor belt mechanical vibration. The pressure signal has a linear correspondence with the steady-state reference; for every 1% increase in the steady-state reference, the pipeline base pressure increases by 1%. The adjustment step size of the pressure signal strictly follows the preset adjustment step size in S3, with a maximum single adjustment not exceeding 3% of the current value of the steady-state reference, achieving smooth adjustment of the pipeline base pressure and ensuring long-term stable control of the base sugar spraying volume.

[0077] The system's fluid piping employs a two-stage independent control structure. A pneumatic proportional valve at the front end of the piping regulates the base pressure of the sugar solution throughout the pipeline, responding to low-frequency steady-state control requirements. A high-speed solenoid valve at the end of each spraying station regulates the transient nozzle opening, responding to high-frequency transient control requirements. The two control stages are independent of each other, receiving pressure and opening signals separately, respectively. The signal transmission links are isolated to prevent coupling interference between the two types of control signals during transmission.

[0078] The system physically superimposes the pipeline's base pressure and the nozzle's transient valve opening at the end nozzle. The final injection flow rate of the sugar solution is jointly determined by the pipeline's base pressure and the valve opening. The base pressure determines the base flow rate range for the sugar solution injection, while the valve opening determines the dynamic adjustment range of the instantaneous flow rate. High-frequency opening signals and low-frequency pressure signals are generated and transmitted independently, and are physically superimposed only at the sugar injection action of the end nozzle. This decouples the feedforward command for response release from the updated steady-state reference, cuts off the cross-oscillation link between feedforward transient action and feedback steady-state adjustment, and consolidates the system's control stability from the execution end.

[0079] The system synchronously mounts the valve action command and actual execution result of each sugar spraying action to the follow-up queue built by S1, using the unique identifier of the corresponding virtual grid as the addressing primary key, to realize full-process traceability of the sugar spraying action and provide complete data support for production process optimization and anomaly investigation.

[0080] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A grain bar physicochemical spraying consistency control system based on multi-sensor fusion, characterized in that, include: The queue module collects the coded pulses of the mixed grain layer conveyor belt, discretizes the physical space of the conveyor belt into a virtual grid, and constructs a follow-up queue with the virtual grid as the primary key. When the feedforward module detects conveyor belt jitter and triggers a feedforward command, it extracts the jitter pulse width, converts the fixed physical delay time of the system into a delay pulse equivalent, compares the jitter pulse width with the delay pulse equivalent to intercept or allow the feedforward command, and records a status label in the corresponding virtual grid. The feedback module obtains the visual sugar spraying thickness of the mixed grain layer and reads the status label corresponding to the virtual grid. If the status label exists and the visual sugar spraying thickness reaches a preset deviation thickness threshold, the visual sugar spraying thickness is removed from the sample pool for updating the steady-state benchmark; otherwise, it is added to the sample pool to update the steady-state benchmark. The execution module superimposes the opening signal generated based on the feedforward command for release with the pressure signal generated based on the updated steady-state reference, and coordinates the control of the end nozzle of the system to perform sugar spraying operation.

2. The system according to claim 1, characterized in that, The queue module is configured as follows when constructing the follow-up queue: Based on a preset unit pulse equivalent, the virtual grid is configured as a first-in-first-out sequence; Identify the physical regions of each segment of the mixed grain layer on the conveyor belt, and establish a lifecycle binding between the physical regions and the uniquely corresponding virtual mesh; The motion detection status, valve action command, and visual detection result are sequentially acquired along the running direction of the conveyor belt, and the motion detection status, valve action command, and visual detection result are attached to the follow-up queue using the identifier of the virtual grid as the addressing primary key.

3. The system according to claim 2, characterized in that, The feedforward module is configured to: extract the jitter pulse width and convert the delayed pulse equivalent. The sum of the computation time of the system, the mechanical action time of the end nozzle, and the fluid flight time is obtained as the fixed physical delay time of the system. Based on the current speed of the conveyor belt and the unit pulse equivalent, the fixed physical delay time is converted into the delay pulse equivalent; Identify pulse intervals that change continuously in the same direction during the motion detection state, and extract their span as the jitter pulse width.

4. The system according to claim 3, characterized in that, The feedforward module is configured as follows: The interception threshold is set based on the delayed pulse equivalent; If the jitter pulse width is less than the interception threshold, the feedforward command is forcibly intercepted, and a transient interference-free label is recorded in the corresponding virtual grid as the status label. If the jitter pulse width is greater than or equal to the interception threshold, the feedforward instruction is allowed to proceed, and a trend-compensated label is recorded in the corresponding virtual grid as the status label. The status label is attached to the follow-up queue and configured for the feedback module to read back, serving as a logical condition for determining whether to remove the corresponding visual spray candy thickness from the sample pool.

5. The system according to claim 4, characterized in that, The feedforward module is also configured to: Construct a hysteresis comparison interval around the interception threshold; When the jitter pulse width falls into the hysteresis comparison interval, the status label of the previous adjacent virtual grid is read from the follow-up queue; The interception or release of the feedforward command is maintained according to the read status tag, and the same status tag is recorded in the corresponding virtual mesh.

6. The system according to claim 5, characterized in that, The feedback module is configured as follows: Backtrack the follow-up queue and read the status label corresponding to the virtual mesh; If the status label is identified as the transient interference-free label or the trend-compensated label, and the visual sugar spraying thickness reaches the deviation thickness threshold, a mask removal operation is triggered. Based on the mask culling operation, the visual candy thickness is removed from the sample pool used to calculate the global steady-state moving average.

7. The system according to claim 6, characterized in that, The feedback module is configured to update the steady-state reference as follows: If it is identified that the corresponding virtual grid does not record the transient interference-free label and the trend-compensated label, and the visual spray candy thickness reaches the deviation thickness threshold, the visual spray candy thickness is incorporated into the sample pool; The steady-state moving average deviation is calculated based on the sample pool filtered by the mask rejection operation. The steady-state reference is updated based on the steady-state moving average deviation and the preset adjustment step size.

8. The system according to claim 7, characterized in that, The execution module is configured to, when superimposing the opening signal and the pressure signal, be: The transient valve opening of the end nozzle is controlled according to the opening signal; Adjust the base pressure of the fluid pipeline connected to the end nozzle according to the pressure signal; The base pressure and the transient valve opening are physically superimposed at the end nozzle to decouple the feedforward command for release from the updated steady-state reference.

9. A method for controlling the consistency of physicochemical spraying of sugar in cereal bars based on multi-sensor fusion, characterized in that, Applied to the system according to any one of claims 1 to 8, the method comprises: Based on the collected coded pulses, the physical space of the conveyor belt is discretized into the virtual grid, and the follow-up queue is constructed using the virtual grid as the primary key; By comparing the extracted jitter pulse width with the converted delayed pulse equivalent, the feedforward command is either forcibly intercepted or allowed to proceed, and the status label is attached to the corresponding virtual grid. Backtrack the state label in the follow-up queue. When the state label exists and the visual spray candy thickness reaches the deviation thickness threshold, remove the visual spray candy thickness from the sample pool used to calculate the steady-state benchmark. Otherwise, merge it into the sample pool to update the steady-state benchmark. The opening signal is generated based on the feedforward command for release, and the pressure signal is generated based on the updated steady-state reference. The opening signal and the pressure signal are superimposed to coordinately control the end nozzle of the system to perform sugar spraying operation.

10. The method according to claim 9, characterized in that, The method further includes: Based on the delayed pulse equivalent, an interception threshold is set, and a hysteresis comparison interval is constructed around the interception threshold. When the jitter pulse width falls into the hysteresis comparison interval, the status label of the previous adjacent virtual grid is traced back, and the interception or release of the feedforward command is maintained according to the status label. Calculate the steady-state moving average deviation of the sample pool and compare the steady-state moving average deviation with a preset dead zone threshold; When the steady-state moving average deviation reaches the dead zone threshold, the steady-state reference is updated based on the steady-state moving average deviation.

Images