Broiler three-dimensional breeding track type whole-process inspection and early warning system and method

Abstract

The application discloses a broiler three-dimensional breeding track type whole-process inspection and early warning system and method, and belongs to the technical field of industrial control system fault diagnosis. The system comprises a track inspection mechanism, a layered environment acquisition assembly, a device visual detection assembly, an industrial communication interface and an edge controller. The method is executed by the edge controller, and each channel profile is collected back and forth and is correlated with each other. The channel time delay is identified by peak displacement, and the sensor time constant is separated to define the upper limit of the cruising speed. Filtering, first-order dynamic inversion and coordinate re-registration are performed to generate layer and section parameter fields. The abnormal score is calculated based on the median and absolute deviation of the median in the same layer as the baseline, and gating and spatial mode discrimination are performed. The actual action state of the execution mechanism is obtained by vision, and is checked with the instruction. The fault source and early warning level are fused and output, and the inspection mode is updated. The application can improve the accuracy of abnormal positioning, shorten the inspection period, suppress false alarms during switching period, and detect the action deviation of the execution mechanism before parameter deterioration.

Description

Technical Field

[0001] This invention belongs to the field of industrial control system fault diagnosis technology, and particularly relates to a track-based full-process inspection and early warning system and method for broiler three-dimensional farming. Background Technology

[0002] This invention relates to an environmental control system and its online monitoring in a broiler chicken integrated farming facility. Such facilities typically have multiple stacked farming layers arranged longitudinally. The environmental control system consists of a programmable logic controller (PLC) and controlled air inlets, exhaust fans, evaporative cooling pads, and heaters. The PLC issues commands to each actuator according to a preset process curve to maintain the temperature, humidity, ammonia, and carbon dioxide levels within the required ranges. These parameters exhibit varying distributions along the longitudinal direction of the facility and along the height of the farming layers; readings from individual measuring points are insufficient to reflect the actual state of the entire facility.

[0003] Currently, there are two main methods for monitoring the above parameters. One method involves installing fixed temperature, humidity, ammonia, and carbon dioxide sensors at regular intervals within the workshop. The sampled values ​​are uploaded to a programmable logic controller (PLC), compared with preset thresholds, and an alarm is triggered. The other method involves laying guide rails within the workshop, with a patrol trolley carrying gas sensors, temperature and humidity sensors, and a camera moving back and forth along the rails. Data is collected, and the measurement points are marked according to the travel distance. The data is then compared with preset thresholds, triggering an alarm. This latter method, using mobile measurement instead of densely distributed measurement points, is closest to the problem addressed by this invention.

[0004] The sensors used to measure ammonia, carbon dioxide, and humidity all exhibit first-order inertia dynamic characteristics. Pump-assisted sampling also introduces transport delays determined by the sampling pipeline volume and flow rate. When the inspection trolley collects data during its movement, the measured values ​​deviate from the true spatial distribution of the measured parameters along the direction of travel, and the measured waveform is simultaneously broadened. This deviation increases with the inspection speed. To control this deviation, current methods either employ intermittent operation with frequent stops and starts or keep the inspection speed very low, correspondingly lengthening the time required for a single round of inspection and hindering timely detection of local anomalies. Once a speed change occurs on the same inspection trajectory, the degree of distortion varies across different sections, making the resulting distribution curves incomparable between sections.

[0005] The dynamic characteristics of sensors also change with use. Electrolyte consumption, dust and feathers adhering to the surface of the breathable membrane, and dust accumulation on the sampling filter can all slow down the response. In existing practices, compensation parameters, if set, are generally determined once at the factory or during installation and commissioning. There is a lack of means to track these changes during operation, and changes in the state of the measurement link itself are not included in the monitoring.

[0006] Within the same workshop, there are inherent systematic differences in temperature and gas concentration across different rearing layers. As the rearing age increases, the baseline for the entire shed gradually rises. Current practices often use fixed absolute thresholds to determine anomalies. These thresholds are too strict in the early stages of rearing and too lenient in the later stages. Furthermore, the same set of thresholds is insufficient to account for different rearing layers. When the overall parameters of the entire shed deviate from the process range, fixed thresholds provide a farm-wide conclusion, which is insufficient to distinguish between the overall situation of mismatch between ventilation volume and current load and equipment anomalies in a specific section or rearing layer.

[0007] Environmental control systems often employ open-loop control for air inlet windows and exhaust fans. The programmable logic controller (PLC) only receives commands from itself, not the actual mechanical operation. Even after issues like linkage disengagement, damper jamming, transmission slippage, or motor phase loss occur, the controller continues its operations according to the commands, often failing to detect the anomaly until environmental parameters significantly deteriorate. When the operating state of the actuators changes, all building parameters require a transition period to reach a new steady state. Current practices tend to generate alarms unrelated to the equipment's status during this period. Furthermore, current methods often only identify the phenomenon of a single parameter exceeding its limit without pinpointing the location of the fault, requiring maintenance personnel to conduct a thorough, end-to-end inspection of the entire building. Summary of the Invention

[0008] To solve the above problems, the technical solution of the present invention is: a track-based full-process inspection and early warning system for broiler three-dimensional farming, comprising: a track inspection mechanism, a layered environmental acquisition component, an equipment visual inspection component, an industrial communication interface, an early warning output unit, and an edge controller, wherein: The track inspection mechanism includes guide rails laid longitudinally along the top or side of the stacked breeding layers in the aquaculture workshop, inspection units that roll in cooperation with the guide rails via traveling wheels, a drive mechanism that drives the inspection units to operate at bidirectional speeds, an incremental encoder for measuring the displacement, position reference components set at equal intervals along the guide rails, and corresponding reading components located in the inspection units. The tiered environmental data acquisition system includes a lifting sensor pod suspended from the inspection unit via a lifting mechanism, with each stopping position corresponding to a sampling height of each aquaculture layer, as well as a gas sampling and detection component, a temperature and humidity sensor, and a wind speed sensor located within the pod. The equipment visual inspection component is located in the inspection unit and is used to independently acquire the actual operating status of the environmental control actuators. The industrial communication interface communicates with the programmable logic controller (PLC) of the environmental control system to obtain the instruction status and timestamps of the actuators. The edge controller serves as the edge control node in an industrial control system. Its inputs are connected to an incremental encoder, a reader, a hierarchical environmental acquisition component, a device vision inspection component, and an industrial communication interface. Its outputs are connected to the drive mechanism, the lifting mechanism, and the early warning output unit. It also includes a channel identification module for sequential data association, a dynamic compensation and spatial re-registration module, an anomaly diagnosis and gating module, an actuator consistency verification module, and an inspection scheduling module. Based on the abnormal diagnosis results and the verification results of the command status and the actual action status, the inspection and scheduling module adjusts the running speed and dwell position of the inspection unit and the stopping position of the lifting sensor pod, outputs the fault source and warning level to the early warning output unit, and sends back the operating mode flag of this round to the channel identification module, forming a closed loop of data collection, diagnosis and scheduling.

[0009] Furthermore, the position reference component is a passive magnet or an RFID tag, and the reader is a corresponding Hall switch or an RFID reader head; the drive mechanism includes a servo geared motor and a friction wheel drive assembly or a synchronous belt drive assembly connected to its output shaft. The traveling wheel rolls with the guide rail surface and bears the weight of the inspection unit. The friction wheel drive assembly or synchronous belt drive assembly is used to transmit traction force without bearing the weight of the inspection unit; the incremental encoder is coaxially connected to the output shaft of the servo geared motor, and its output end is connected to the dynamic compensation and spatial re-registration module together with the output end of the reader. The output signal of the reader is used to perform absolute position correction on the cumulative displacement of the incremental encoder.

[0010] Furthermore, the gas sampling and detection assembly includes a sampling port equipped with a dust filter, a sampling pipeline, a gas sensor, and a sampling pump. The gas sensor includes an electrochemical ammonia sensor and a non-dispersive infrared carbon dioxide sensor. The gas chambers of both are connected in parallel to the sampling pipeline via their respective flow restrictors, so that the two gas channels have their own independent branch volumes and branch flow rates. The sampling pump is connected to the downstream side of the two gas chambers and makes the sampling gas flow unidirectionally from the sampling port to the gas chambers. The temperature and humidity sensor is directly exposed to the ventilation window of the lifting sensing pod without passing through the sampling pipeline. The wind speed sensor is a hot-film wind speed sensor, and its output end is connected to the actuator consistency verification module.

[0011] Furthermore, the device's visual inspection component includes a visible light camera fixed to a two-degree-of-freedom gimbal, as well as a supplementary light and a dust cover. The two-degree-of-freedom gimbal is positioned to the side of the inspection unit and its rotation and pitch angles are controlled by the edge controller. The edge controller stores a calibration mapping between the orbital coordinates of each environmental control actuator and the corresponding gimbal angle. When the inspection unit is stationary, the optical axis of the visible light camera is aligned with the junction of the window panel and window frame of the air inlet window or with the air outlet of the exhaust fan according to the calibration mapping. The image output terminal of the visible light camera is connected to the actuator consistency verification module.

[0012] Furthermore, the industrial communication interface is an industrial Ethernet interface, which reads in the opening command of the air inlet window, the start / stop and frequency commands of the exhaust fan and their command timestamps, and sends back the fault source location number and warning level; the three input terminals of the actuator consistency verification module are respectively connected to the equipment visual inspection component, the industrial communication interface and the wind speed sensor, and its output terminals are respectively connected to the inspection scheduling module and the warning output unit; the warning output unit includes a field audible and visual alarm and a host computer monitoring terminal; the edge controller is also connected to a local storage unit, which stores the original measurement sequence, the compensated sequence, the calibration parameters and the supplementary inspection queue of each channel.

[0013] This invention also provides a track-based, end-to-end inspection and early warning method for broiler three-dimensional farming, including the following steps: S1, During the period when the command status of the environmental control actuator remains unchanged, each track segment runs at the cruise speed and there are identifiable spatial changes in environmental parameters along the track, the control inspection unit completes adjacent outbound and return acquisitions at the same constant speed. For each environmental parameter channel, the uncompensated outbound and return profiles of the channel are normalized and cross-correlated. The total equivalent time delay of the channel is identified by the spatial displacement corresponding to the peak value of the correlation coefficient. After deducting the sampling and transport delay of the channel, its first-order time constant is obtained, and the cruise speed upper limit is set according to the equivalent time constant after compensation for each channel. S2, after a single-pass acquisition is completed, the sensor output of each channel is filtered and the first-order dynamic inversion is performed according to the first-order time constant of each channel to eliminate the measurement tail caused by sensor inertia. The track coordinates of each sampling point are corrected according to the instantaneous speed and the sampling and transport delay of the channel. The cumulative displacement error of the encoder is corrected with the position reference component. Then, the layer and section parameter fields of each channel are generated by resampling according to the breeding layer and track section. S3 calculates the anomaly score for each channel's parameter field based on the median and median absolute deviation of each section of the same breeding layer. After the command state is switched, the overall alarm is gated within the set time determined by the net volume of the breeding workshop and the effective exhaust volume. The suspected anomaly location and suspected anomaly type are output according to the distribution pattern of the anomaly score along the breeding layer direction and along the track direction. S4, the control inspection unit stays and retests at suspected abnormal locations and corresponding actuators. The equipment vision inspection component independently obtains the actual action status of the actuator, performs consistency verification with the command status of the same control cycle, and merges it with the result of S3. It outputs the fault source and warning level, and updates the next round of patrol, detailed inspection or stay operation mode and supplementary inspection queue for each track section according to the comprehensive value of the abnormality score of each channel.

[0014] Furthermore, the environmental parameter channels include an ammonia channel, a carbon dioxide channel, a temperature channel, and a humidity channel, to... Indicating the channel number, S1 includes: S11, when the command state remains unchanged within a time window consisting of a one-way trip and an adjacent return trip, when each track segment runs at cruising speed within the time window, when the duration of the time window is not greater than the quotient of the net volume of the aquaculture workshop and the current effective exhaust volume, and when the standard deviation of the outbound measurement sequence along the track coordinates is not less than three times the resolution of the channel sensor, a calibration permission signal is issued; when no time window meeting the above conditions appears within the set calibration period, a dedicated calibration round trip is arranged in which each track segment runs at cruising speed; S12, control the inspection unit to operate at a constant speed The outward journey is completed along the increasing direction of the orbital coordinates, and then the return journey is completed at the same constant speed. For the ... The original measurements of the channel without dynamic compensation are resampled at the same track coordinate intervals to obtain outbound and return profiles. For each candidate spatial displacement, the outbound profile is translated along the track coordinates by that displacement, and a normalized cross-correlation coefficient is calculated between the outbound profile and the return profile within their effective overlap interval. The value range of the candidate spatial displacement does not exceed one-quarter of the profile length, and the number of resampling points within the effective overlap interval is not less than a set lower limit for the number of overlap points. The spatial displacement corresponding to the peak value of this coefficient is taken. and that peak value ; S13, decomposed according to the following formulas, yields the... The first-order time constant of the channel sensor: ; In the formula, For the first The total equivalent delay of the channel; The first obtained for S12 Peak displacement of the channel; The constant inspection speed during calibration; For the first The first-order time constant of the channel sensor; For the first The sampling and transport delay of the channel is obtained by measuring the pure delay time when a standard gas of known concentration is introduced into the sampling port of the channel during the installation and commissioning phase to form a step input. It is corrected according to the ratio of the sampling branch flow rate at the calibration time to the current sampling branch flow rate. The temperature and humidity channels that do not pass through the sampling pipeline are taken as zero. S14, in Not less than the relevant threshold , Greater than zero and Greater than Time to determine the first The channel identification was valid this time, and the results obtained this time will be used to further improve the channel. The saved value is updated after being weighted and averaged with the channel's saved value at a ratio of 1:9; if it is determined to be invalid, the original saved value is retained and the channel's calibration invalidation flag is set; when the saved value is greater than twice the channel's calibration value during the installation and commissioning phase, a channel degradation warning is output; where, The first obtained for S12 Peak value of channel normalized cross-correlation coefficient; The relevant threshold is determined by the lower bound of the distribution of peak values ​​obtained from multiple consecutive round trips during the installation and commissioning phase under normal operating conditions. S15, set the maximum cruising speed using the following formula: ; In the formula, This is the maximum cruising speed. To achieve the required spatial resolution, the longitudinal length of a cage unit is taken. The response completion coefficient is set to 3, which corresponds to the first-order element completing 95.0% of the step response. For the first The equivalent time constant of the channel after S2 compensation is determined by applying a known step excitation to the channel during the installation and commissioning phase, taking the time it takes for the compensated output of the channel to change from the initial value to 63.2% of the final value, and then recalibrating it after replacing the channel sensor or dust filter. This indicates that the larger of the channel values ​​within the curly braces is selected.

[0015] Furthermore, in S2, sampling and orbital coordinate correction are performed in real time during the movement of the inspection unit, while filtering, inversion, and reregistration are performed in batches on the sequence of a single journey after the single-journey acquisition is completed. S2 includes: S21, for the first The channel uses the sampling period of that channel. Synchronously record sensor output and orbital coordinates With instantaneous velocity When the reader detects a position reference, the track coordinates are set to the calibrated position of that reference, and the scaling factor is updated by the ratio of the distance between two adjacent position references to their corresponding encoder pulse increments, used for real-time conversion of track coordinates within the interval after that reference; the wind speed sensor does not participate in the acquisition during travel, and its measurement values ​​are acquired in the dwell state of S41; where, For the first The sampling period of a channel is the data output period of the sensor in that channel. The sampling sequence number; For the first The corrected orbital coordinates of each sampling point; For the first The instantaneous velocity of each sampling point; S22, after a single-trip data collection is completed, the first data collection point within that single trip... The channel sequence is first averaged once along the increasing sampling number direction and then once along the decreasing sampling number direction. The two ends of the sequence are extended according to the endpoint values. The window length of the moving average is the odd number of points corresponding to the quotient of one-tenth of the first-order time constant of the channel and the sampling period of the channel, rounded to the nearest integer, to obtain the filtered sequence. Then, a first-order dynamic inversion is performed according to the following formula, and the inversion result is limited to the range of the sensor of the channel: ; In the formula, For the first Channel 1 Compensation value for each sampling point; and The first Channel 1 The and the first The filtered value of each sampling point; The first obtained for S1 The channel's first-order time constant; For the first The sampling period of the channel; filtering once in both directions to make the net phase lag of the single program sequence zero; S23, assign each compensation value back to its actual sampling position according to the following formula, and then... Using the segment length as the unit and each stop position of the lifting sensor pod as the aquaculture layer number, the arithmetic mean of the compensation values ​​for pods falling into the same segment of the same aquaculture layer is taken to generate the first... Channel layer and section parameter fields : ; In the formula, For the first Channel 1 The orbital coordinates of each sampling point after reregistration; For the first The orbital coordinates of each sampling point are obtained from S21; The sign function is +1 when the inspection unit runs along the direction of increasing track coordinates and -1 when running along the direction of decreasing track coordinates; For the first The absolute value of the instantaneous velocity at each sampling point; For the first Sampling and transport delay of the channel; For the first The installation offset of the channel sampling port relative to the positioning reference point of the inspection unit is positive in the direction of increasing track coordinates; The segment length is the same as the spatial resolution required in S15. For the first Channel 1 Aquaculture layer The parameter field values ​​of the section; For breeding layer number, The numbers are segment numbers, all of which are dimensionless; the first-order dynamic inversion is performed in the time domain while the position translation is performed point by point according to the instantaneous velocity. The parameter fields of each channel obtained in this step are independent of the inspection speed.

[0016] Furthermore, S3 includes: S31, for the first For each culture layer in the channel, the median of the parameter field values ​​of each segment of that culture layer is taken as the robust baseline of that culture layer. The median absolute deviation of the culture layer is taken as the absolute value of the absolute difference between the parameter field values ​​of each section and the robust baseline. Then, calculate the anomaly score for each section according to the following formula, and take the larger of the absolute values ​​of the anomaly scores of each channel in the same section of the same breeding layer as the comprehensive score for that section of the breeding layer. : ; In the formula, For the first Channel 1 Aquaculture layer Abnormal scoring for the section; The parameter field values ​​obtained from S2; For the first Channel 1 A robust baseline for the culture layer; For the first Channel 1 The absolute deviation of the median in the culture layer; 0.6745 is the conversion constant for converting the absolute deviation of the median to the standard deviation, taken from the values ​​of the quartiles of the standard normal distribution; This indicates that the larger of the values ​​within the curly braces is selected. For the first Resolution of the channel sensor; For the first Aquaculture layer The comprehensive score of the section; the absolute value of the abnormal score is not less than the abnormality threshold. The passage, the aquaculture layer, and the corresponding section were immediately identified as abnormal. Take 3; S32, starting from the moment the instruction state changes, within the set time determined by the following formula. No overall alarm will be output within this period: ; In the formula, For setting time; To fine-tune the multiplier, during the installation and commissioning phase, a step increase in the number of operating units was applied to the exhaust fans. The time taken for each channel parameter to reach 95% of the new steady state was recorded, along with the time taken at that moment. and The ratio of the quotients is taken as the larger of the ratios obtained from each channel; This refers to the net volume of the breeding workshop; To determine the effective exhaust volume, the sum of the number of operating units of each type of exhaust fan and the rated air volume of that type of exhaust fan under the current working static pressure is taken. S33, for each channel, determine the abnormal score according to the spatial distribution pattern: the abnormal scores of each culture layer in the same segment on the same channel have the same sign and their absolute values ​​are not less than 1. When an abnormality is detected in a section, the section number is output and the section is considered a suspected abnormality. The absolute value of the abnormality score in at least three consecutive sections on the same channel within the same aquaculture layer must be no less than [amount missing]. The absolute values ​​of the anomaly scores for the remaining aquaculture layers in the same section were less than [a certain value]. When the culture layer is suspected of having an interlayer environmental anomaly, the culture layer number and section range are output; if the robust baseline of each culture layer on the same channel simultaneously exceeds the process setting range, and the maximum absolute value of the channel anomaly score is less than [missing information], then [missing information]. If the overall operating condition deviates, no output will be given within the validity period of the gate signal in S32; this step outputs the suspected abnormal location and suspected abnormal type, and the specific fault source is determined by S4.

[0017] Furthermore, S4 includes: S41, the overall score in a certain section is not less than When the value is twice that of the previous value, or when S33 determines that the section is suspected to be an abnormal air supply path or an abnormal inter-floor environment, the inspection unit is controlled to stay in the section. During the stay, the lifting sensor pod is controlled to stop at each floor, and the steady-state remeasurement value of each channel and the wind speed value at the stay position are collected. The stay time is the product of the larger of the response completion coefficient and the equivalent time constant after compensation of each channel. S42, during the installation and commissioning phase, at least five air inlet window opening commands, evenly distributed from fully closed to fully open, are issued sequentially. After each command is executed, the visible light camera acquires an image and measures the angle between the window panel and the window frame, establishing a calibration mapping between the command opening and the window panel angle. During operation, edge detection is performed on the image of the air inlet window at the stationary position, the straight line segment between the window panel and the window frame is extracted, and its angle is measured. The actual opening is obtained by piecewise linear interpolation on the calibration mapping. , The unit is percentage; S43, acquire image sequences of the exhaust fan outlet area at a set frame rate, and calculate normalized inter-frame motion features at two non-integer multiple frame intervals respectively; during the installation and commissioning phase, measure the stop feature value and the operation feature value at the same two frame intervals in both the stop and operation states of the exhaust fan; when the operation feature value at the same frame interval is greater than the stop feature value, and the motion features measured at the operation in both frame intervals are less than the arithmetic mean of the stop feature value and the operation feature value corresponding to that frame interval, it is determined that the exhaust fan blades are not rotating; S44, Obtain the opening degree of the air intake window command within the same control cycle as the image acquisition time. In conjunction with the start / stop commands of the exhaust fan, and The absolute value of the difference is greater than the consistency threshold. If, at any time, or when the exhaust fan start / stop command is "run" but S43 determines that the blades are not rotating, it is determined that the actuator has a deviation fault where the command and actual action are inconsistent. In the formula, The opening degree of the air intake window command issued by the programmable logic controller; The consistency threshold is determined by combining the measured mechanical backlash of the air inlet window actuator linkage mechanism with the repeatability measured value of the image angle measurement in S42 using the square root. Then, the output is fused according to the following correspondence: when a suspected abnormal air supply path is determined and the air inlet window in that section is determined to have a deviation fault, output a fault in the air inlet window actuator of that section; when a suspected abnormal air supply path is determined, the air inlet window in that section is not determined to have a deviation fault, and the wind speed at the residence position obtained in S41 is lower than the median wind speed at each residence position in that aquaculture layer, output a blocked ventilation path in that section; when an overall operating condition is determined to be deviated and there is an exhaust fan whose start / stop command is "run" but whose blades are not rotating, output a transmission or electrical fault in that exhaust fan. S45, based on the comprehensive score, sets the operating mode for each section in the next round of inspections: comprehensive score less than The section operates at the maximum cruising speed, not less than Sections less than twice the speed of the cruise speed are inspected at a speed determined by the ratio of the cruise speed limit and the response completion coefficient, and image acquisition is encrypted. Sections not less than twice the speed of the cruise speed are kept in S41. When the scheduling time of a single round of inspection exceeds the set value, the sections to be handled in this round are selected from the highest to the lowest comprehensive score, and the sequence number of the delayed section is written into the supplementary inspection queue. The sections in the supplementary inspection queue are arranged in the next round of inspection with priority over sections with the same score and are re-inspected within no more than two rounds of inspection. When the same fault source recurs in at least two consecutive rounds of inspection, a level 2 warning is output. When an action deviation fault is determined in S44, a level 1 warning with priority over the level 2 warning is output.

[0018] The beneficial effects of this invention are as follows: This invention performs online identification of the dynamic characteristics of the measurement channel during the round-trip operation of the inspection unit. The outbound and return journeys traverse the same spatial distribution in opposite directions, resulting in opposite but equal distortion directions in the measured values. The relative displacement between the two profiles is proportional to the total equivalent time delay of the measurement channel, a relationship free of pre-assumed coefficients. The edge controller obtains this relative displacement using normalized cross-correlation and then subtracts the sampling transport delay obtained from standard gas step calibration, thus separating the sensor's own first-order time constant. This identification is completed during normal production inspections without requiring shutdown, the introduction of additional calibration gas, or interruption of environmental control. The identification results are recursively updated based on three criteria: correlation peak value, displacement sign, and time delay magnitude. Random fluctuations in a single identification are suppressed, while slow parameter drift can still be tracked. The slowdown in response caused by electrolyte consumption, dust adhering to the breathable membrane, and dust accumulation on the sampling filter is manifested as a continuous increase in the time constant in the identification results. This invention directly uses this increase as a criterion for the degradation of the measurement channel itself and prompts maintenance accordingly, so that the reliability of the measurement link can be continuously monitored.

[0019] This invention eliminates the two error components of the measurement channel in the time and spatial domains, respectively. The sensor's first-order inertia broadens the measured waveform and shifts its centroid backward. The edge controller first filters the single program sequence along both directions of the sampling sequence number to suppress high-frequency noise without introducing additional phase lag. Then, it performs dynamic inversion based on the identified first-order time constant, restoring the waveform's broadening and shift. The transport delay of the sampling path does not change the waveform but only causes an overall translation. The edge controller shifts the compensation value back to its true sampling position based on the instantaneous velocity of each sampling point and the transport delay of that channel, and eliminates the error of the encoder's cumulative displacement using a position reference element set along the guide rail. After the above processing, the reported abnormal location corresponds to the actual distribution location of the measured parameter, allowing maintenance personnel to reach the section where the abnormality is located based on the reported location. The parameters used in the inversion calculation are all time-related and independent of the inspection speed. The speed-related part is completely isolated in the position translation term and calculated point by point. The parameter fields obtained by segmented speed changes at different speeds, and even within the same inspection trip, converge to the same spatial distribution, making the measurement results of each segment comparable. Compensation reduces the equivalent time constant of the measurement channel. Based on this, the invention quantitatively converts the required spatial resolution into the upper limit of the cruise speed. Without relaxing the resolution requirements, the inspection unit can pass through normal sections faster, and the time required for a single round of inspection is correspondingly shortened, leading to more timely detection of anomalies.

[0020] This invention uses the median of each segment in the same breeding layer and the same cycle as the baseline, and the absolute deviation of the median as the measure of dispersion to calculate the anomaly score. The inherent inter-layer differences in stratified breeding are automatically subtracted when comparing within the same layer. The baseline, which gradually increases with the age of the breeders, is also synchronously followed. The anomaly threshold does not need to be adjusted separately with the breeding layer and age, thus avoiding underreporting in the early stages of breeding and widespread over-limiting in the later stages. The median and the absolute deviation of the median remain bounded even when some samples are contaminated; a few strongly anomalous segments will not raise the anomaly threshold, and the detection capability does not decrease when multiple anomalous segments appear in the same breeding layer. After the actuator state switch, there is a transition process in the overall building parameters. This invention determines the setting time based on the net volume of the breeding workshop and the current effective exhaust volume, and gates the overall alarm within this time. The setting time varies with the real-time exhaust volume, consistent with the actual ventilation process. Alarms during the transition period are suppressed, while actual faults are still output after the setting time. The present invention further distinguishes abnormal scores based on the distribution pattern along the direction of the breeding layer and along the direction of the track. Each breeding layer deviates synchronously from the air supply path pointing to that position at the same longitudinal position. The same breeding layer deviates from the equipment pointing to that breeding layer in multiple consecutive sections. There is no synchronous deviation of the ventilation volume pointing to the current load in the entire space structure. The scope of the abnormality is narrowed down to the specific section or the specific breeding layer.

[0021] This invention utilizes a vision inspection component mounted on the inspection unit to independently acquire the actual operational status of the environmental control actuators and compares it with the command status issued within the same control cycle. The actual opening degree of the air inlet window is obtained by reverse lookup using a multi-point calibration mapping established during the installation and commissioning phase, absorbing the nonlinearity of the linkage mechanism. The rotational state of the exhaust fan is determined by calculating motion characteristics at two non-integer multiple frame intervals and jointly judging them, suppressing misjudgments caused by the coupling of the blade cycle and sampling interval. The moving vision channel provides common status feedback for multiple dispersed actuators, closing control loops that previously lacked position and speed feedback. Deviations in actuator movement are detected before environmental parameters accumulate to exceed limits, eliminating the need for individual feedback components for each actuator. After the verification results and spatial pattern discrimination conclusions are integrated, the output is no longer a phenomenon of a parameter exceeding the limit, but a fault source with a location number, narrowing the scope of on-site troubleshooting to specific locations. The diagnostic results are then sent back to the inspection dispatcher. Sections with high anomaly scores are inspected more thoroughly in the next round at a lower speed or stopped for retesting, while normal sections continue to pass at cruising speed. Limited inspection time is reallocated to locations requiring attention. Stopping for retesting obtains steady-state values ​​under zero-speed conditions, providing independent confirmation of the in-process measurement conclusions and eliminating occasional anomalies in in-process measurements. Sections that exceed the scheduling timeout are added to the supplementary inspection queue and given priority in the next round, ensuring traceability of the inspection coverage. There are data and control dependencies among the above steps. Online identification provides parameters for dynamic compensation, dynamic compensation provides a comparable parameter field for anomaly scores, anomaly scores specify the parts to be verified for consistency checks, and the instruction status read by consistency checks is the trigger signal for transition process gating. The dispatch results then return to online identification to provide round-trip data that meets the trigger conditions. If any step is missing, the functions of the remaining steps cannot be achieved independently. The effectiveness of this invention is built upon the overall technical solution. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the composition structure of the broiler three-dimensional farming track-type full-process inspection and early warning system of the present invention. Detailed Implementation

[0023] The technical solutions of this invention will be clearly and completely described below. In addition, the forms of the various structures described in the following embodiments are merely illustrative. This invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0024] This implementation method targets broiler chicken integrated farming workshops and their environmental control systems, such as... Figure 1As shown, the aquaculture workshop has multiple rows of stacked aquaculture layers arranged longitudinally, typically 2 to 4 layers, with a longitudinal length of 80m to 120m. The environmental control system includes a programmable logic controller (PLC) and controlled components such as air inlets, exhaust fans, evaporative cooling pumps, and heaters. The PLC issues commands to each actuator according to a preset process curve. However, there is usually no mechanical position or speed feedback between the PLC and the air inlets / exhaust fans. Therefore, the PLC only knows the command status, not the actual operating status of the machinery. If a linkage disengages, a damper jams, a belt slips, or a motor phase is lost, the control loop will fail without being detected. Simultaneously, the distribution of ammonia, carbon dioxide, temperature, and humidity within the workshop varies along both the longitudinal and vertical directions. Fixed-point sensors cannot provide a full-length distribution, while mobile measurements are limited by the dynamic characteristics of the sensors. This embodiment constructs the inspection device as an edge control node of the industrial control system. While collecting environmental parameters, it identifies the dynamic characteristics of the measurement channel itself online, verifies the consistency between the instructions and actual actions of the actuator, and adjusts the inspection method based on the diagnostic results.

[0025] The inspection and early warning system of this embodiment includes a track inspection mechanism, a layered environmental acquisition component, an equipment visual inspection component, an industrial communication interface, an early warning output unit, and an edge controller.

[0026] The track inspection mechanism includes a guide rail, inspection unit, drive mechanism, incremental encoder, position reference component, and reading component. The guide rail is laid longitudinally along the aquaculture workshop, above or to the side of the stacked aquaculture layers. It runs the entire length of the track and has mechanical limit components and proximity switches for deceleration at both ends. Its function is to constrain the movement of the inspection unit to one dimension, so that the position of any measurement point can be represented by a single track coordinate. This is a prerequisite for subsequent registration of measurement values ​​with spatial positions. The inspection unit rolls on the guide rail via traveling wheels, which also bear the weight of the inspection unit. The drive mechanism includes a servo geared motor and a friction wheel transmission assembly or synchronous belt transmission assembly connected to the motor's output shaft. This transmission assembly transmits traction force without bearing the weight of the inspection unit, thus separating load-bearing and traction, preventing the traction component from experiencing increased wear due to also serving as a load-bearing component. The servo geared motor allows the inspection unit to run at varying speeds along the guide rail and to stop at any track coordinate, providing the execution conditions for subsequent speed adjustment and fixed-point stopping based on diagnostic results. The incremental encoder is coaxially connected to the output shaft of the servo geared motor and is used to measure the running displacement.

[0027] Position reference components are fixedly installed along the guide rail at equal intervals, preferably 3m to 10m. They can be passive magnets or RFID tags. The reader is located in the inspection unit and corresponds to the installation position of the position reference components. When used with passive magnets, a Hall switch is used; when used with RFID tags, an RFID reader is used. The output of the incremental encoder and the output of the reader are connected to the dynamic compensation and spatial re-registration module of the edge controller. In dusty environments, friction wheels or timing belts may slip. Relying solely on the encoder's accumulated pulses will result in a mileage error that increases with the travel distance. The reader provides an absolute position each time it passes a position reference component, thereby correcting the accumulated displacement. The error in the track coordinates is thus limited to between two adjacent position reference components and no longer accumulates along the entire length.

[0028] The stratified environmental acquisition system includes a lifting mechanism, a lifting sensor pod, a gas sampling and detection component, temperature and humidity sensors, and wind speed sensors. The lifting mechanism is a winch mechanism connected to the lifting sensor pod via a steel wire rope and equipped with a lifting encoder to measure the pod's height. The lifting sensor pod is suspended below the inspection unit, and its stopping position corresponds one-to-one with the sampling height of each aquaculture layer, with a preferred stopping repeatability accuracy better than 20mm. Thus, measurement values ​​can be obtained layer by layer from the same track coordinates. The measurement results form a two-dimensional distribution along the track direction and along the layer height direction, rather than a single curve at a single height, allowing the inherent inter-layer differences in stacked aquaculture to be observed separately.

[0029] The gas sampling and detection assembly includes a sampling port equipped with a dust filter, a sampling pipeline, a gas sensor, and a sampling pump. The sampling port is located in the lifting sensor pod, and the dust filter is used to prevent feathers and dust from entering the pipeline. The gas sensors include an electrochemical ammonia sensor and a non-dispersive infrared (NDIR) carbon dioxide sensor. The gas chambers of the two sensors are connected in parallel to the sampling pipeline via their respective flow restrictors. The sampling pump is connected downstream of the two gas chambers, allowing the sampling gas flow to flow unidirectionally from the sampling port to the gas chambers. The sampling branch flow rate is preferably 0.3 L / min to 1.0 L / min, and the sampling tube inner diameter is preferably 3 mm to 6 mm. The gas chambers are connected in parallel instead of in series. This avoids the electrochemical ammonia sensor consuming ammonia during measurement, which would affect the downstream reading. Furthermore, it ensures that each gas channel has an independent branch volume and flow rate, resulting in independent transport delays, facilitating separate identification and compensation. The temperature and humidity sensor is directly exposed to the ventilation window of the lifting sensor pod without passing through the sampling pipeline, thus eliminating pipeline transport delays in its measurements. The wind speed sensor is a hot-film type, and its output is connected to the actuator consistency verification module of the edge controller. The hot-film type wind speed sensor measures the magnitude of the airflow speed but does not contain directional information. If it is collected during the movement of the inspection unit, the result is the combined speed of the airflow and the movement of the inspection unit, which cannot be restored by scalar addition and subtraction. Therefore, wind speed measurement is arranged to be carried out when the inspection unit is stationary, and the result is used to assist in the judgment of the state of the air supply path.

[0030] The equipment's visual inspection components include a visible light camera fixed to a two-degree-of-freedom (DOF) pan-tilt head, supplementary lighting, and a dust cover. The two-DOF pan-tilt head is positioned to the side of the inspection unit, and its rotation and pitch angles are controlled by an edge controller. The pan-tilt head's repeatability is preferably better than 0.5°, the camera resolution is preferably no less than 1920×1080, and the frame rate is preferably no less than 25 frames / s. The edge controller stores a calibration mapping between the orbital coordinates of each environmental control actuator and the corresponding pan-tilt head angle. When the inspection unit is stationary, the visible light camera's optical axis is aligned with this calibration mapping, pointing towards the junction of the air inlet window's panel and frame, or towards the exhaust fan's outlet. The supplementary lighting, controlled by the edge controller, ensures image quality in low-light conditions inside the building. The dust cover protects the lens. The visible light camera's image output is connected to the actuator consistency verification module. The role of the equipment vision inspection component is to provide a common and movable status feedback channel for multiple actuators that are distributed in a distributed manner, so that the actuator loop that originally had no position feedback and speed feedback can be closed, without having to install sensors on each actuator individually.

[0031] The industrial communication interface is an industrial Ethernet interface, which connects to the PLC of the environmental control system. It reads data including the opening command of the air inlet window, the start / stop and frequency commands of the exhaust fan, and their timestamps. The returned data includes the fault location number and warning level. The command timestamp ensures that the command status and the image acquisition time are aligned to the same control cycle, preventing misjudgments caused by changes in commands before and after image acquisition. The warning output unit includes a local audible and visual alarm and a host computer monitoring terminal. The edge controller is also connected to a local storage unit to store the original measurement sequences, compensated sequences, calibration parameters, and re-inspection queues for each channel.

[0032] The edge controller serves as the edge control node in an industrial control system. Its inputs are connected to an incremental encoder, a reader, a hierarchical environmental acquisition component, a device vision inspection component, and an industrial communication interface, respectively. Its outputs are connected to the drive mechanism, the lifting mechanism, and the early warning output unit, respectively. The edge controller incorporates a channel identification module for sequential data association, a dynamic compensation and spatial reregistration module, an anomaly diagnosis and gating module, an actuator consistency verification module, and an inspection scheduling module. The channel identification module outputs dynamic parameters for each measurement channel, which are then sent to the dynamic compensation and spatial reregistration module. The dynamic compensation and spatial reregistration module outputs two-dimensional parameter fields for each channel's layers and segments, which are sent to the anomaly diagnosis and gating module. The anomaly diagnosis and gating module outputs suspected anomaly locations and types, which are sent to the actuator consistency verification module. The actuator consistency verification module's first input is connected to the equipment's visual inspection component, its second input to the industrial communication interface, and its third input to the wind speed sensor. Its outputs are connected to the inspection scheduling module and the early warning output unit, respectively. Based on the anomaly diagnosis results and the verification results of the command status and actual action status, the inspection scheduling module adjusts the inspection unit's operating speed and dwell position, as well as the stopping position of the lifting sensor pod. It outputs the fault source and early warning level to the early warning output unit and sends back the current cycle's operating mode flag to the channel identification module. This feedback allows the channel identification module to determine whether each segment of the current cycle is operating at cruising speed, thus forming a closed loop between data acquisition, diagnosis, and scheduling.

[0033] Based on the above system, the inspection and early warning method of this embodiment is executed by the edge controller and includes four steps from S1 to S4. S1 identifies the dynamic characteristics of the measurement channel online and sets the upper limit of the cruise speed; S2 performs dynamic compensation and track coordinate re-registration on the moving measurement results and generates a parameter field; S3 performs robust anomaly scoring, transient process gating and spatial mode determination on the parameter field; S4 performs consistency verification between the instruction and actual action of the actuator, determines the fault source and updates the inspection mode for the next round.

[0034] The environmental parameter channels described below include ammonia channel, carbon dioxide channel, temperature channel, and humidity channel, in order to... This indicates the channel number. The dynamic characteristics, sampling paths, and dimensions of the sensors in each channel are different. For example, the response time of an electrochemical ammonia sensor is usually on the order of tens of seconds, followed by an NDIR carbon dioxide sensor, and then a temperature and humidity sensor. Since the temperature and humidity channels do not pass through sampling pipelines, the calculations of S1 to S3 are performed separately for each channel, and each channel has its own set of dynamic parameters and parameter field.

[0035] S1 comprises five sub-steps, S11 to S15. S11 is the determination of calibration conditions. The edge controller reads the command status of the environmental control actuator via the industrial communication interface and issues a calibration permission signal when all four of the following conditions are met simultaneously: First, the command status remains unchanged within a time window consisting of one outbound trip and the immediately following return trip; second, each track segment operates at cruising speed within this time window; third, the duration of this time window is not greater than the quotient of the net volume of the aquaculture workshop and the current effective exhaust volume; fourth, the standard deviation of the outbound measurement sequence along the track coordinates is not less than three times the resolution of the channel sensor. The above four conditions correspond to the prerequisites for the relevant calculations to be valid. Maintaining a stable concentration field is essential, as changes in the number of fans or the opening of small windows immediately alter the overall concentration distribution. The ratio of net volume to effective exhaust volume represents the nominal ventilation time of the workshop. Within a window shorter than this time, the concentration field variation is limited, and the outbound and return journeys can be considered two observations of the same spatial distribution. Identical speeds ensure that the spatial distortions of the outbound and return journeys are equal in magnitude but opposite in direction. The standard deviation constraint guarantees the existence of spatial fluctuations on the profile that can be used for correlation calculations. If the concentration is nearly uniform along the track, the correlation function will not exhibit a clear peak. If no time window satisfying the above four conditions occurs within the set calibration period, the edge controller arranges a dedicated calibration round trip where each track segment runs at cruising speed to prevent calibration from being impossible if abnormal sections exist in the workshop for an extended period.

[0036] S12 is for round-trip data acquisition and related calculations. The edge controller controls the inspection unit to operate at a constant speed. The journey begins in the direction of increasing orbital coordinates, followed by the return journey at the same constant speed, and then... The uncompensated raw measurements of the channel are resampled at the same orbital coordinate intervals to obtain the outbound and return profiles. It is crucial to use the uncompensated raw values ​​here; if compensated values ​​are used, the time delay to be identified has been eliminated, and the relevant calculations will lose their target. Subsequently, for each candidate spatial displacement... Translate the outbound profile along the orbital coordinates. Then, the normalized cross-correlation coefficient is calculated between the return profile and the return profile within their effective overlap interval, and the calculation formula is as follows.

[0037] ; In the formula, For the first Channel displacement measurement Normalized cross-correlation coefficient (dimensionless) over time. The candidate spatial displacement (m); The orbital coordinates are in meters (m). For the first The outbound profile of the channel in orbital coordinates The original measured value at the location (in the same unit as the parameter being measured in this channel); For the first The return profile of the passage in orbital coordinates The original measured value at the location (in the same unit as the parameter being measured in this channel); It is the arithmetic mean of the measurements of each resampling point in the effective overlap interval of the outgoing profile under the candidate displacement (in the same unit as the measured parameter of the channel). It is the arithmetic mean of the measured values ​​of each resampling point in the effective overlap interval of the return profile under the candidate displacement (in the same unit as the measured parameter of the channel). This represents the summation of all resampling points within the effective overlapping interval under the candidate displacement; This represents the square root operation. Both the mean and the sum of squares are expressed as... Calculations are performed within the effective overlapping interval of the variation. If a fixed mean from the entire profile is used instead, the results will differ. The corresponding summation interval lengths are inconsistent, and the correlation coefficient will have a variable effect. Boundary deviation of changes. The range of candidate spatial displacement values ​​is preferably no more than 1 / 4 of the profile length, and the number of resampling points within the effective overlap interval is preferably no less than 1 / 2 of the total number of points in the profile. If the number of overlap points is insufficient, the candidate displacement will not participate in peak selection. The spatial displacement corresponding to the peak value is denoted as This peak value is denoted as The correlation function can be interpolated using a three-point parabola fit near its peak, making... The resolution is better than one resampling interval.

[0038] S13 represents time delay decomposition. The measurement channel consists of a first-order inertial element of the sensor connected in series with a pure time delay element of the sampling path. The inspection unit moves along the direction of increasing orbital coordinates at a speed... During travel, the measured values, relative to the actual spatial distribution, translate along the direction of increasing orbital coordinates. The translation amount is the product of velocity and the equivalent time delay of the channel. On the return journey, the direction of travel is reversed, and the same mechanism causes the measured values ​​to translate along the direction of decreasing orbital coordinates, with the same translation amount. Therefore, on the same orbital coordinate axis, the relative displacement between the outbound and return profiles is twice the product of the equivalent time delay of the channel and velocity. (The outbound profile is translated...) The correlation coefficient is related to the return profile and is within the range of [missing information]. The peak value is obtained when the relative displacement is equal to that value, i.e. Equal to 2 times The total equivalent time delay is obtained by producting the total equivalent time delay with the sampling transport delay. When the spatial variation of the measured quantity is dominated by low-frequency components, the equivalent time delay of the first-order inertial element is approximately equal to its time constant. The equivalent time delay of the pure delay element is its delay time. When the two are connected in series, their equivalent time delays are added together. Therefore, subtracting the sampling transport delay from the total equivalent time delay yields the sensor's first-order time constant. The corresponding calculation formula is as follows.

[0039] ; In the formula, For the first The total equivalent delay of the channel (s); The first obtained for S12 Peak displacement of the channel (m); The constant inspection speed (m / s) during calibration. For the first The first-order time constant (s) of the channel sensor; For the first The sampling and transport delay of the channel (s). During the installation and commissioning phase, the calibration is determined by a standard gas step calibration. That is, a standard gas of known concentration is introduced into the sampling port of the channel to form a step input. The pure delay time between the moment the step is applied and the moment when the sensor output begins to deviate from the baseline is taken as the calibration value. During operation, the calibration is corrected according to the change in the flow rate of the sampling branch. The calculation formula is as follows.

[0040] ; In the formula, For the first The current sampling transport delay (s) of the channel; For the first The pure delay time (s) obtained by calibrating the channel during the installation and commissioning phase; For calibration The flow rate (mL / s) of the sampling branch of this channel at that time; This represents the current flow rate (mL / s) of the sampling branch in this channel. This correction is based on the fact that the time required for gas to pass through a fixed-volume pipe and gas chamber is inversely proportional to the volumetric flow rate. For temperature and humidity channels that do not pass through a sampling pipe, Take zero, its It is also zero. The quotient of the sampling pipeline volume and the sampling pump flow rate can be used as... The initial value estimate is used, but the mixing of the filter dead volume with the airflow in the air chamber will cause the estimate to deviate from reality, so the step calibration value is used as the standard.

[0041] S14 is for validity determination and parameter update. The edge controller... Not less than the relevant threshold , Greater than zero and Greater than Time to determine the first The channel identification was valid this time. The lower bound of the peak value distribution obtained by repeated round trip calibration under normal operating conditions during the installation and commissioning phase is determined, and it is preferably taken as 0.8 to 0.9 in engineering. A value greater than zero corresponds to the physical fact that the translation directions of the outbound and return journeys are opposite. If the calculated peak displacement is negative, it indicates that the correlation peak comes from noise or that the concentration field has drifted during the round trip. Greater than Guarantee that the result obtained by S13 Positive and negative time constants have no physical meaning and will cause the inversion of S22 to diverge. If this identification is valid, the results obtained this time... The stored value is updated after being weighted and averaged with the channel's stored value at a ratio of 1 to 9. The calculation formula is as follows.

[0042] ; In the formula, For the first time saved after the update First-order time constant of the channel (s); The first one saved before the update First-order time constant of the channel (s); This is the first one obtained from S13. First-order time constant of the channel (s); To update the weights (dimensionless). The relationship between the equivalent average frequency and the frequency is as follows.

[0043] ; In the formula, To update the weights (dimensionless); The equivalent average calibration number (times). When the exponentially weighted recursion and the simple average of equal length are equivalent in autocorrelation time, the above equation is satisfied. Take... It was 9 o'clock. A value of 0.2 is equivalent to averaging the results of the most recent nine identifications. This suppresses random fluctuations in a single identification while still allowing for the tracking of slow parameter drift. When an identification is deemed invalid, the original saved value is retained, and the calibration invalidation flag for that channel is set to prevent subsequent compensation from being contaminated by erroneous parameters. Electrolyte consumption, dust and feather contamination of the breathable membrane surface, and dust accumulation on the filter element all increase the first-order time constant. Therefore, when the saved value exceeds twice the calibration value during the channel's installation and commissioning phase, a channel degradation warning is output. The parameter drift itself is thus transformed into a diagnostic criterion for the measurement channel itself, eliminating the need for separate detection methods.

[0044] S15 sets the upper limit of cruising speed. To achieve the set value for spatial resolution along the track direction, the time taken for the inspection unit to traverse a segment should not be less than the time required to reach the set response completion rate after channel compensation. The inspection unit uses speed... Through length of The time taken for the section is and The first-order process begins from the application of the step jump. Completed after doubling the time constant The upper limit of cruising speed is obtained by taking the response of the former as not less than that of the latter and selecting the slower response for each channel. The calculation formula is as follows.

[0045] ; In the formula, This is the upper limit of cruising speed (m / s); To achieve the required spatial resolution, the longitudinal length (m) of a cage unit is selected, preferably 1m to 3m; The response completion coefficient (dimensionless) is preferably set to 3, which corresponds to 95.0% of the first-order element completing the step response. If the resolution requirement is relaxed, it can be set to 2.3, which corresponds to 90% of the step response completing the step response. For the first The equivalent time constant (s) of the channel after S2 compensation; This indicates that the larger of the channel values ​​within the curly braces is selected. Calibration is performed during the installation and commissioning phase. This involves applying a known step excitation to the channel and taking the time it takes for the compensated output to change from its initial value to 63.2% of its final value. Recalibration is then performed after replacing the channel sensor or the dust filter. Because the inversion of S22 changes the equivalent time constant from... and The sum decreases to With the same spatial resolution requirements, the upper limit of cruising speed is increased accordingly, and the increase factor is the ratio of the two, thus shortening the time required for single-wheel inspection.

[0046] S2 comprises three sub-steps, S21 to S23. Sampling and orbital coordinate correction are performed in real-time during the movement of the inspection unit, while filtering, inversion, and reregistration are performed in batches on the sequence after a single-trip acquisition is completed. This division is made because the zero-phase filtering in S22 requires processing once in each of the two directions along the sampling sequence number, and the reverse processing must be performed after the entire sequence acquisition is completed. It is a non-causal operation and should not be confused with real-time filtering.

[0047] S21 is for synchronous sampling and orbit coordinate correction. The edge controller... The channel uses the sampling period of that channel. Synchronously record sensor output and orbital coordinates With instantaneous velocity In the formula, For the first The sampling period (s) of the channel is the data output period of the sensor in that channel. The gas channel is generally 1s, while the temperature and humidity channel can be shorter than 1s. The sampling sequence number (dimensionless); For the first Corrected orbital coordinates (m) of each sampling point; For the first The instantaneous velocity (m / s) at the sampling point. The orbital coordinate correction method is to detect the instantaneous velocity (m / s) at the sampling point. When a position reference is used, the track coordinates are set to the calibration position of the reference, and the scaling factor is updated according to the ratio of the distance between two adjacent position references to the corresponding encoder pulse increment. This is used for real-time conversion of track coordinates in the interval after the reference. The calculation formula is as follows.

[0048] ; In the formula, is the scaling factor (m / pulse); The distance between adjacent reference components (m); The sequence number of the position reference component (dimensionless); To read the first The encoder's cumulative pulse count (pulses) at each position reference point; To read the first The encoder's cumulative pulse count (pulses) at each position reference point; Accumulate the current number of encoder pulses. The current sampling point's orbital coordinates (m) are given. The scaling factor is obtained from the most recent interval's actual measurements, thus incorporating proportional errors caused by friction wheel wear and slippage. The wind speed sensor does not participate in data acquisition during travel; its measurements are collected in the stationary state of S41.

[0049] S22 represents zero-phase filtering and first-order dynamic inversion. After a single-pass acquisition is completed, the first step is to analyze the data within that single pass. The channel sequence is averaged once along the direction of increasing sampling number, and then the result is averaged again with the same window length along the direction of decreasing sampling number. The sequence is extended at both ends according to the endpoint values ​​to avoid distortion caused by incomplete windows at the endpoints. The phase responses of the two filters are equal in magnitude but opposite in sign, and the combined phase response is zero, thus not introducing any additional time lag. This is crucial for the subsequent inversion; if the filtering itself introduces lag, the time constant used for the inversion will no longer match reality. The window length for the moving average is determined by the following formula.

[0050] ; In the formula, For the first Window length (points) for channel moving average; This indicates that the value within the square brackets is rounded to the nearest integer. The first obtained for S1 First-order time constant of the channel (s); For the first The sampling period of the channel (s). When the value is taken according to the above formula, the equivalent time constant of the filter is approximately... The value is 1 / 10, much smaller than the time constant to be compensated. High-frequency noise is suppressed while the dynamic information on which the inversion is based is not weakened. Furthermore, the odd window length ensures that the center of the moving average falls on the sampling point. The filtered sequence is denoted as... The dynamic characteristics of the sensor can be described by a first-order inertial element, and its time-domain equation is: Multiplying by the time derivative of the output and adding the output equals the input. Solving for the input yields the result that the input equals the output plus the product of the time constant and the time derivative of the output. Applying backward difference discretization to the derivative, we obtain the inversion equation as follows.

[0051] ; In the formula, For the first Channel 1 The compensation value for each sampling point (in the same unit as the parameter being measured in this channel); For the first Channel 1 The filtered value of each sampling point (in the same unit as the parameter being measured in this channel); For the first Channel 1 The filtered value of each sampling point (in the same unit as the parameter being measured in this channel); For the first The first-order time constant (s) of the channel sensor; For the first The sampling period (s) of the channel. The two terms on the right-hand side of the above equation have the same dimensions; the second term is the product of time and the rate of change of the parameter with respect to time, still in the dimension of the parameter itself. The inversion result is limited upper and lower according to the range of the sensor in that channel to prevent noise spikes from exceeding the range after differential amplification. After the above processing, the measurement tail caused by sensor inertia is eliminated, and the concentration peak is no longer broadened and shifted towards the direction of travel of the inspection unit.

[0052] S23 involves orbital coordinate reregistration and parameter field generation. The inversion in S22 eliminates the first-order inertial element in the time domain; the remaining pure time delay element does not change the waveform but only causes an overall translation, i.e., the... The value recorded at each sampling time actually corresponds to the value before that time. The gas entering the sampling port, and the position of the inspection unit at that time, which is the product of the current coordinate and the speed of retreat along the direction of travel and the delay, and taking into account the installation offset of the sampling port relative to the positioning reference point of the inspection unit, the re-registration formula is as follows.

[0053] ; In the formula, For the first Channel 1 The orbital coordinates (m) of each sampling point after reregistration; For the first The orbital coordinates (m) of each sampling point obtained from S21; The sign function is positive 1 when the inspection unit moves in the direction of increasing track coordinates and negative 1 when moving in the direction of decreasing track coordinates (dimensionless). For the first The absolute value of the instantaneous velocity at each sampling point (m / s); For the first The sampling and transport delay of the channel (s); For the first The installation offset (m) of the channel sampling port relative to the positioning reference point of the inspection unit is taken as positive in the direction of increasing track coordinates and is measured from the assembly dimensions. Subsequently, using... The track is divided into several sections according to the section length. Each stopping position of the lifting sensor pod is designated as the aquaculture layer number. The arithmetic mean of the compensation values ​​for pods falling into the same section of the same aquaculture layer is taken to obtain the number of sections. Two-dimensional parameter fields of channel layers and sections In the formula, For the first Channel 1 Aquaculture layer The parameter field value of the section (in the same unit as the measured parameter of the channel); For aquaculture layer number (dimensionless); The segment number (dimensionless); The segment length is the same as the spatial resolution required in S15 (m).

[0054] The inversion formula of S22 only contains and The two time parameters do not include velocity. The velocity-related part is concentrated in the translation term of S23 and calculated based on the instantaneous velocity point by point. Therefore, regardless of the speed at which the inspection unit operates, or even if the speed is varied in segments within the same inspection run, the resulting parameter field converges to the same spatial distribution. Decoupling the parameter field from the inspection speed is a prerequisite for subsequent speed adjustment based on diagnostic results. Without the above processing, the degree of measurement distortion in each segment will differ at different speeds, making it impossible to compare segments and establish a robust baseline.

[0055] S3 includes three sub-steps, S31 to S33.

[0056] S31 involves establishing a robust baseline and calculating anomaly scores. For the... For each aquaculture layer in the channel, the median of the parameter field values ​​of each segment of the aquaculture layer is taken as the robust baseline of the aquaculture layer, and the median absolute deviation of the aquaculture layer is taken as the median absolute deviation of the aquaculture layer. The median absolute deviation is denoted as MAD and its calculation formula is as follows.

[0057] ; In the formula, For the first Channel 1 Robust baseline of the culture layer (in the same units as the parameters measured in this channel); For the first Channel 1 Median absolute deviation of the culture layer (in the same unit as the parameter measured in this channel); This indicates that the median value of each segment of the channel and the culture layer within the curly braces is taken; The parameter field value obtained from S2 (in the same unit as the measured parameter of this channel); This indicates taking the absolute value. Anomaly scores are then calculated based on this, using the following formula.

[0058] ; In the formula, For the first Channel 1 Aquaculture layer Anomaly score for the section (dimensionless); The parameter field value for this channel, this aquaculture layer, and this section (in the same unit as the measured parameter of this channel); This serves as the robust baseline for the culture layer in this channel (in the same units as the measured parameters in this channel). 0.6745 is the median absolute deviation of the culture layer in this channel (in the same unit as the measured parameter of this channel); 0.6745 is the conversion constant (dimensionless) for converting the median absolute deviation to the standard deviation. Its value is taken from the upper quartile of the standard normal distribution. For a sample that follows a normal distribution, the ratio of the median absolute deviation to the standard deviation is equal to the upper quartile. Therefore, dividing the median absolute deviation by this constant gives a robust estimate of the standard deviation, which is equivalent to multiplying the numerator by this constant. This indicates that the larger of the two values ​​enclosed in curly braces is selected. For the first The resolution of the channel sensor (in the same unit as the measured parameter of that channel) serves to prevent the denominator from becoming zero when the parameter field has almost no fluctuations. Anomalies in that channel, culture layer, and section are determined to be abnormal when the absolute value of the anomaly score is not less than the anomaly threshold. The anomaly threshold is preferably set to 3, corresponding to a two-sided probability of approximately 0.0027 for samples following a normal distribution. The baseline and dispersion are constructed using the median and MAD, with a breakdown point of 50%. This means that the estimator remains bounded when less than half of the sections are contaminated, and a few strongly anomalous sections will not raise the anomaly threshold. This property is not present when constructing using the arithmetic mean and standard deviation. The baseline is taken from sections within the same culture layer and cycle. The inherent inter-layer system differences in stratified culture are automatically subtracted, and the baseline automatically shifts upward with the age of the rearing animals, eliminating the need to look up tables by age. Finally, the larger of the absolute values ​​of the anomaly scores for each channel within the same section of the same culture layer is taken as the comprehensive score, calculated as follows.

[0059] ; In the formula, For the first Aquaculture layer The overall score of the section (dimensionless). Anomaly scores (dimensionless) are given for each channel in this section of the aquaculture layer. This indicates that the larger of the channel values ​​within the curly braces is selected. This indicates the absolute value. The overall score is used for the inspection and scheduling of S45.

[0060] S32 is a transition process gating. When the number of fans, the opening degree of the small windows, or the status of the heaters change, the overall parameters of the aquaculture facility will transition from one steady state to another. During this period, the baselines of each aquaculture layer will synchronously deviate from the process setting range. If alarms are triggered without differentiation, false alarms unrelated to the equipment status will occur. When the workshop is considered an equivalent fully mixed container, the response of the parameters to a step change in ventilation volume is approximately a first-order process. Its time constant is the quotient of the net volume of the workshop and the effective exhaust volume, i.e., the nominal air exchange time. Reaching 95% of the new steady state requires approximately three times this time constant. Considering that the actual airflow is not ideally fully mixed, the on-site calibrated setpoint is used instead of the fixed three times. The formula for calculating the setpoint time is as follows.

[0061] ; In the formula, The settling time is in seconds. For the set multiple (dimensionless); The net volume of the breeding workshop (m³) 3 ); Effective exhaust volume (m³) 3 / s). During the installation and commissioning phase, calibration is performed by applying a step change in the number of operating units to the exhaust fan, recording the time taken for each channel parameter to reach 95% of the new steady state, and taking that time and the time taken at that moment. and The ratio of the quotients is taken as the larger of the ratios obtained for each channel. In longitudinally ventilated workshops with good airflow organization, this ratio is generally close to 3. When the airflow is short-circuited or there are dead zones, this ratio increases. The formula for calculating the effective exhaust volume is as follows.

[0062] ; In the formula, Effective exhaust volume (m³) 3 / s); For the first Number of operating exhaust fans (units); For the first The rated air volume (m³) of the type of exhaust fan under the current working static pressure 3 / s), the value of which is taken from the air volume and static pressure curve of the fan sample; This refers to the model number (dimensionless) of the exhaust fan. This represents the summation over all types of exhaust fans. The calculation begins from the moment the command state changes. It does not output a general alarm, but only records the data. The settling time varies with the number of operating fans in real time. The settling time is shorter when more fans are running and longer when fewer fans are running, which is consistent with the actual ventilation process and is better than using a fixed delay.

[0063] S33 is for spatial pattern determination. The edge controller determines the spatial distribution pattern of the abnormal scores for each channel. When the abnormal scores of all aquaculture layers in the same channel within the same segment are the same and their absolute values ​​are not less than the anomaly threshold, the segment is determined to be a suspected air supply path anomaly and the segment number is output. This is based on the synchronous deviation of each layer at that longitudinal position, pointing to the air supply path at that position rather than the equipment of a particular layer. When the absolute values ​​of the abnormal scores of at least three consecutive segments in the same channel of the same aquaculture layer are not less than the anomaly threshold, while the absolute values ​​of the abnormal scores of other aquaculture layers in the same segment are less than the anomaly threshold, the aquaculture layer is determined to be a suspected inter-layer environmental anomaly and the aquaculture layer number and segment range are output. This is based on the deviation extending along the track direction without crossing layers, pointing to the water line, feed line, or inter-layer guide component of that layer. If the robust baseline of each aquaculture layer on the same channel synchronously exceeds the process setting range, and the maximum absolute value of the anomaly score for that channel is less than the anomaly detection threshold, it is determined that the overall operating condition has deviated. At this time, there is no spatial structure in the entire field, indicating a mismatch between ventilation volume and current load rather than a local fault. This conclusion will not be output within the validity period of the S32 gating signal. The above determination results in the suspected anomaly location and suspected anomaly type. Spatial distribution alone is insufficient to determine the specific equipment; the specific fault source is determined by S4.

[0064] S4 includes five sub-steps, S41 to S45.

[0065] S41 is for dwell time retesting. When the edge controller's comprehensive score for a certain section is not less than twice the anomaly detection threshold, or when S33 determines that the section is suspected of having an abnormal air supply path or a suspected abnormal inter-floor environment, the control inspection unit will dwell in that section. During the dwell time, the control will stop the lifting sensor pod layer by layer, collecting the steady-state retest values ​​of each channel and the wind speed value at the dwell position. The dwell time is calculated as follows.

[0066] ; In the formula, The dwell time is in seconds. The response completion coefficient (dimensionless) is the same value as S15. For the first The equivalent time constant (s) of the channel after S2 compensation; This indicates that the larger of the values ​​for each channel within the curly braces is selected. During the dwell period, the speed of the inspection unit is zero, and there is no spatial translation. The steady-state remeasurement value constitutes an independent confirmation of the measurement conclusion during travel, and occasional anomalies caused by residual errors from moving measurements are eliminated here.

[0067] S42 measures the actual opening of the air inlet window. The air inlet window actuator typically drives the window panel via a linkage mechanism. The relationship between the actuator stroke and the window panel angle is generally not linear, and the percentage command issued by the PLC is not equal to the percentage of the window panel angle. Therefore, a linear conversion between fully closed and fully open points is not used. During installation and commissioning, at least five air inlet window opening commands, evenly distributed from fully closed to fully open, are sequentially issued to the PLC. After each command is executed, a visible light camera captures an image and measures the angle between the window panel and the window frame. This establishes a calibration mapping between the commanded opening and the window panel angle, which is then stored in the edge controller. During operation, edge detection is performed on the image of the air inlet window at the stationary position. The straight line segment between the window panel and the window frame is extracted, and its angle is measured. The actual opening is then obtained by piecewise linear interpolation on the calibration mapping, as shown in the following formula.

[0068] ; In the formula, The actual opening (%) of the air inlet window as measured by the image. The angle (°) between the window panel and the window frame is measured during operation. To determine the dimensionless index of the calibration point in the calibration mapping, take the value that makes... Falling and The sequence number between them; For the first Command opening degree (%) of each calibration point; For the first Command opening degree (%) of each calibration point; For the first The angle (°) between the window panel and the window frame at each calibration point; For the first The angle (°) between the window panel and the window frame at each calibration point. The denser the calibration points, the better the piecewise linear interpolation approximates the nonlinearity of the linkage mechanism. In engineering, 5 to 9 calibration points are sufficient.

[0069] S43 determines the actual rotation state of the exhaust fan. The edge controller acquires image sequences of the exhaust fan outlet area at a set frame rate, and calculates normalized inter-frame motion features at two non-integer multiples of each other. This feature is the average of the absolute values ​​of the grayscale differences of corresponding pixels within the same region of interest in two adjacent frames, divided by the average grayscale value of that region to eliminate the influence of illumination changes. The region of interest is abbreviated as ROI. During the installation and commissioning phase, the stop and operation feature values ​​are measured at the same two frame intervals for both the stop and operation states of the exhaust fan. If the operation feature value at the same frame interval is greater than the stop feature value, and the motion features measured at both frame intervals are less than the arithmetic mean of the stop and operation feature values ​​corresponding to that frame interval, it is determined that the exhaust fan blades are not rotating. The reason for using two frame intervals is that the image brightness changes caused by the rotation of the blades have the same period as the frequency of the blades passing through. When the frame interval is exactly an integer multiple of this period, the blade positions in the two frames being compared are close, the motion characteristics are small, and it is difficult to distinguish them from the stationary state. When two frame intervals that are not integer multiples of each other are used, the probability of both falling into this condition at the same time is greatly reduced. The condition that the running characteristic value is greater than the stationary characteristic value is used to confirm that the calibration itself is effective. If the two are close, it indicates that the ROI selection, exposure time, or supplementary lighting conditions are inappropriate and need to be reset before calibration.

[0070] S44 is for consistency verification and fault source fusion. The edge controller acquires the air inlet window opening command and the exhaust fan start / stop command within the same control cycle as the image acquisition. When the absolute value of the difference between the actual opening and the commanded opening is greater than the consistency threshold, or when the exhaust fan start / stop command is running but S43 determines that the blades are not rotating, it is determined that the actuator has a deviation fault where the command and actual action are inconsistent. In the formula, The opening degree (%) of the air intake window command issued by the PLC; The consistency threshold (percentage points) is determined by combining the measured mechanical hysteresis of the air inlet window actuator linkage mechanism with the measured repeatability of the image angle measurement in S42 using the square root of the sum of their squares. In engineering practice, this threshold typically falls between 5 and 15 percentage points, with the specific value determined by on-site measurements and not pre-set. Subsequently, the output is fused according to the corresponding relationship. If a suspected abnormality in the air supply path is identified and the air inlet window in that section is determined to have a deviation fault, a fault in the air inlet window actuator of that section is output. If a suspected abnormality in the air supply path is identified, but the air inlet window in that section is not determined to have a deviation fault, and the wind speed at the residence position obtained in S41 is lower than the median wind speed at each residence position in that aquaculture layer, a blocked ventilation path is output. If an overall operating condition deviation is identified and there is an exhaust fan with a start / stop command indicating operation but whose blades are not rotating, a transmission or electrical fault in that exhaust fan is output. After the above integration, the output is no longer a phenomenon of a parameter exceeding the limit, but a fault source with a location number. The scope of on-site troubleshooting is narrowed from the entire building to a single section or piece of equipment. Deviations in the actuator's operation are detected before environmental parameters deteriorate significantly, thus closing the actuator loop that previously had no feedback.

[0071] S45 is for inspection scheduling and tiered early warning. The edge controller sets the operating mode for each section in the next round of inspection based on the comprehensive score. Sections with a comprehensive score less than the anomaly threshold operate at the upper limit of the cruise speed; sections with a comprehensive score not less than the anomaly threshold but less than twice it operate at a detailed inspection speed and increase the density of image acquisition. The formula for calculating the detailed inspection speed is as follows.

[0072] ; In the formula, To examine the speed (m / s) in detail; The maximum cruising speed (m / s) set for S15; The response completion coefficient (dimensionless) is the same value as S15. When running according to the above formula, the spatial resolution of this segment increases to... and Sections with a comprehensive score not less than twice the anomaly threshold are retained according to S41. When the scheduling time of a single round of inspection exceeds the set value, sections to be handled in this round are selected according to the comprehensive score from largest to smallest, and the sequence number of the postponed section is written into the supplementary inspection queue. Sections in the supplementary inspection queue are given priority over sections with the same score in the next round of inspection, and the re-inspection is completed within no more than 2 rounds of inspection, so the postponed sections will not be missed for a long time. When the same fault origin occurs repeatedly in no less than 2 consecutive rounds of inspection, a level 2 warning is output. When an action deviation fault is determined in S44, a level 1 warning is output, which takes priority over the level 2 warning. Since the proportion of abnormal sections in the entire workshop is limited, the above classification keeps the overall workshop inspection cycle basically unchanged, while the measurement confidence is concentrated on abnormal sections.

[0073] There are clear data and control relationships between S1 to S4. The first-order time constant identified by S1 is the sole source of the filter window length and inversion coefficients in S22; the sampling transport delay identified by S1 is the sole source of the translation amount in S23; the cruise speed upper limit set by S1 is the benchmark for each speed setting in S45; the parameter field output by S2 is the sole input for the baseline and score calculation in S31. If reregistration is not performed, the abnormal position itself has been translated, and the spatial structure on which S33 is based will be destroyed; the spatial mode conclusion of S33 determines which actuators S42 and S43 verify, avoiding indiscriminate cycling of all actuators; the command status read by S44 is the trigger signal for gating in S32; the operating mode flag of S45 is sent back to S11, providing round-trip data that meets the trigger conditions for the next identification. The above relationships are interconnected, and the function of the remaining links cannot be obtained independently if any link is missing.

[0074] This implementation can also have several alternatives. The position reference can be replaced by an RFID tag or reflective target instead of a passive magnet, and the reader can be replaced by an RFID reader or photoelectric sensor accordingly. The transmission component can be replaced by a friction wheel drive instead of a synchronous belt drive or a rack and pinion drive. The gas sensor can add a hydrogen sulfide channel or a dust channel, and the added channel is identified in the same way as its first-order time constant and sampling transport delay and incorporated into the parameter field. The zero-phase filtering in S22 can be replaced by a bidirectional moving average instead of a bidirectional first-order low-pass filter, and the method for determining the window length can be replaced by the method for determining the time constant. The robust dispersion in S31 can be replaced by the median absolute deviation instead of the interquartile range, and the conversion constant can be replaced accordingly, while the remaining steps remain unchanged. The piecewise linear interpolation in S42 can be replaced by monotonic cubic interpolation at the calibration point. The above alternatives and the remaining steps of S1 to S4 can be used in any combination, and the data correlation between the steps after combination is the same as described above.

[0075] Example 1. The net volume of a certain broiler chicken integrated farming workshop is 4500m³. 3 The structure is 100m long, 15m span, and 3m high, with four stacked aquaculture layers. It is equipped with four negative pressure exhaust fans with impeller diameters of 1380mm, each with a rated air volume of 12.2m³ / h under the current operating static pressure. 3 / s, this airflow rate is taken from the airflow and static pressure curve of the fan sample. The guide rail runs the entire length of the walkway, with an effective travel of 96m. The position reference components use passive magnets arranged at 5m intervals, and the reading components are Hall switches. The lifting sensor pod has four stops, corresponding to the sampling height of the four floors. The gas sampling and detection component has a sampling tube 1.2m long and 4mm inner diameter. The gas chambers of the electrochemical ammonia sensor and the non-dispersive infrared carbon dioxide sensor are connected in parallel via their respective flow limiters. The flow rate of each branch is 0.3L / min, and the total volume of the pipeline and gas chamber is approximately 17mL. The sampling pump is connected downstream of the gas chamber. The visible light camera of the equipment's visual inspection component has a resolution of 1920×1080 and a frame rate of 25 frames / s. Four environmental parameter channels are used. The channels are, in order: ammonia channel, carbon dioxide channel, temperature channel, and humidity channel.

[0076] The operation process of this embodiment is as follows. In S1, S11 determines the calibration conditions and takes a calibration period of 24h; S12 completes round-trip acquisition at a constant speed and performs cross-correlation, the candidate spatial displacement range is taken as 1 / 4 of the profile length, the lower limit of the number of overlapping points is taken as 1 / 2 of the total number of points in the profile, and the peak position is obtained by three-point parabolic fitting interpolation; S13 decomposes to obtain the first-order time constant and sampling transport delay of each channel; S14 takes a correlation threshold of 0.85 and an equivalent average calibration number of 9; S15 takes a spatial resolution of 2m and a response completion coefficient of 3. In S2, S21 synchronously samples according to the data output period of each channel and uses a magnet for orbital coordinate correction, and the gas channel sampling period is taken as 1s; S22 performs forward and reverse bidirectional filtering and first-order dynamic inversion; S23 performs coordinate re-registration and generates a parameter field according to a 2m segment length, the entire length is divided into 48 segments, and the peak position of the parameter field is obtained by three-point parabolic fitting interpolation without being limited by segment discretization. In S3, S31 calculates the robust baseline, median absolute deviation, and anomaly score, with an anomaly threshold of 3; S32 calculates the settling time based on the number of operating wind turbines and performs gating; S33 determines the spatial mode. In S4, S41 retests the section with a comprehensive score of not less than 6; S42 establishes a calibration mapping between the command opening and the window panel angle based on 5 calibration points; S43 takes two frame intervals of 2 frames and 5 frames; S44 performs consistency verification and fault source fusion; S45 schedules the wind turbines according to three levels: cruise, detailed inspection, and dwell, with a maximum single-round scheduling time of 20 minutes, a maximum of 2 rounds for continuous confirmation, and a maximum of 2 rounds for supplementary inspection completion.

[0077] Example 2. The difference from Example 1 is that the parameter values ​​are closer to the lower limit. The aquaculture workshop is a three-layer wire mesh floor-raising facility. Position reference components are arranged at 3m intervals. The sampling tube inner diameter is 3mm, the flow rate of each of the two branches is 0.5L / min, the total volume of the pipeline and air chamber is approximately 11mL, the spatial resolution is 1m, the response completion coefficient is 2.3, 9 calibration points are used for calibration mapping, and the correlation threshold is 0.80. The remaining steps are the same as in Example 1.

[0078] Example 3. The difference from Example 1 is that the parameter values ​​are closer to the upper limit. The breeding workshop is two-story, the position reference components are arranged at 10m intervals, the sampling tube inner diameter is 6mm, the flow rate of each of the two branches is 0.3L / min, the total volume of the pipeline and air chamber is approximately 36mL, the spatial resolution is 3m, and the relevant threshold is 0.90. The remaining steps are the same as in Example 1.

[0079] Example 4. The difference from Example 1 lies in the use of a combination of several alternative implementation methods. The position reference component is replaced with an RFID tag, and the reader is correspondingly replaced with an RFID reader head; the friction wheel drive assembly of the drive mechanism is replaced with a synchronous belt drive assembly; a hydrogen sulfide channel is added to the gas sampling and detection assembly, which identifies its first-order time constant and sampling transport delay and incorporates them into the parameter field in the same manner as the other gas channels, increasing the number of environmental parameter channels to 5; the median absolute deviation in S31 is replaced with the interquartile range, and the robust estimate of the standard deviation is correspondingly taken as the quotient of the interquartile range and 1.349; the piecewise linear interpolation in S42 is replaced with monotonic cubic interpolation at the calibration point. The remaining steps are the same as in Example 1.

[0080] Comparative Example 1. The difference from Example 1 is that S22 does not perform first-order dynamic inversion, but only retains forward and reverse bidirectional filtering. The coordinate re-registration in S23 is performed as usual, and the inspection speed is the same as in Example 1.

[0081] Comparative Example 2. The difference from Example 1 is that S23 does not perform coordinate reregistration, that is, the translation term is set to zero, and the first-order dynamic inversion of S22 is performed as usual. The inspection speed is the same as that of Example 1.

[0082] Comparative Example 3. This comparative example serves as a blank control closest to the prior art. The differences from Example 1 are: S22 does not perform filtering or first-order dynamic inversion; S23 does not perform coordinate reregistration; S3 uses a fixed absolute threshold for each channel instead of robust scoring and spatial pattern determination; the ammonia threshold is 15 ppm; and S4 does not perform actuator consistency verification. In the spatial positioning and speed-variable comparability test, this comparative example operates at the same inspection speed as Example 1; in the inspection cycle test, this comparative example operates at the upper speed limit obtained by substituting the uncompensated equivalent time constant into the calculation formula of S15.

[0083] Comparative Example 4. The difference from Example 1 is that in S31, the arithmetic mean is used instead of the median, and the standard deviation is used instead of the absolute deviation of the median. The outlier threshold is still 3, and the rest of the steps are the same.

[0084] Comparative Example 5. The difference from Example 1 is that S42 to S44 are not set, that is, the actual operating state of the actuator is not obtained, and the fault determination is based only on the spatial mode conclusion and parameter limit violation in S33.

[0085] This manual contains eight appendices. Table 1 shows the dynamic parameter calibration and online identification results for each environmental parameter channel; Table 2 shows the upper limit of cruising speed and single-trip inspection time for each example; Table 3 shows the spatial positioning error; Table 4 shows the profile comparability under variable speed conditions; Table 5 shows the inter-layer and full-cycle consistency of anomaly detection sensitivity; Table 6 shows the impact of the proportion of abnormal sections in the same layer on the anomaly detection score; Table 7 shows the synthesis of the actuator consistency threshold and the opening measurement results; and Table 8 shows the fault detection time and fault source location results.

[0086] Experiment 1: Online Identification of Channel Dynamic Parameters and Upper Limit of Inspection Speed. This experiment uses the methods of Examples 1 to 4 and Comparative Example 3. First, standard gas of known concentration is introduced into the sampling ports of each channel. A step input is formed by switching using a three-way solenoid valve. The time taken for the output to change from the initial value to 63.2% of the final value is taken as the reference value of the first-order time constant. The time from the application of the step input to the output starting to deviate from the baseline is taken as the calibration value of the sampling transport delay. The temperature and humidity channels are switched using a constant temperature and humidity chamber to form the step input. Then, under normal workshop conditions, online identification is performed 20 times consecutively according to S11 to S14. Outliers are identified and removed using conventional statistical methods. Finally, the upper limit of the cruising speed is calculated according to formula S15, and the single-trip inspection time is measured over an effective travel distance of 96m. The experimental results are shown in Tables 1 and 2.

[0087] Table 1. Dynamic parameter calibration and online identification results for each environmental parameter channel: First-order time constant reference value / s 58.0 24.0 6.5 11.0 Sampling transport delay calibration value / s 3.4 3.1 0 0 Online identification of mean / s 56.8 23.4 6.3 10.6 Online identification of standard deviation / s 2.6 1.3 0.5 0.8 Relative deviation / % −2.1 −2.5 −3.1 −3.6 Peak mean of correlation coefficient 0.94 0.96 0.92 0.91 Effective identification rate / % 90 95 85 85 Equivalent time constant after compensation / s 7.4 3.6 1.2 1.9 The ratio of equivalent time constants before and after compensation 8.1 7.4 5.3 5.6

[0088] Table 2 shows the maximum cruising speed and single-trip inspection time for each example: Spatial resolution / m 2 1 3 2 2 Response completion coefficient 3 2.3 3 3 3 The equivalent time constant of the larger of the two factors is taken in seconds. 7.4 7.4 7.4 7.4 60.2 Maximum cruising speed (m / s) 0.090 0.059 0.135 0.090 0.011 Single inspection time / min 17.8 27.1 11.9 17.8 145.5

[0089] In Table 1, the first-order time constant reference value and the sampling transport delay calibration value are obtained from step calibration. The online identification mean and standard deviation are statistics from 20 identifications. The relative deviation is the ratio of the difference between the online identification mean and the reference value to the reference value. The peak mean of the correlation coefficient reflects the similarity of the round-trip profile. The effective identification percentage is the ratio of the number of times the results were determined to be effective according to S14 to the total number of times. The equivalent time constant after compensation is obtained by step calibration of the compensated output. The ratio before and after compensation is the ratio of the sum of the first-order time constant and the sampling transport delay to the equivalent time constant after compensation. In Table 2, the equivalent time constant with the larger value is the quantity corresponding to the channel with the largest value in S15. For Examples 1 to 4, the value is 7.4s after compensation, and for Comparative Example 3, it is 60.2s before compensation. As shown in Table 1, the relative deviation between the online identification value and the step calibration value of the four channels does not exceed 4%, and the standard deviation does not exceed 8% of the mean. The equivalent time constant after compensation is reduced by 5.3 to 8.1 times compared to before compensation.

[0090] As shown in Table 2, the upper limits of cruising speed for Examples 1 to 3 are 0.090 m / s, 0.059 m / s, and 0.135 m / s respectively, due to differences in spatial resolution and response completion coefficient. All of these values ​​fall within the relationship determined by the calculation formula in S15, and the single-trip inspection times are 17.8 min, 27.1 min, and 11.9 min respectively. Although Example 4 adds a hydrogen sulfide channel, the equivalent time constant after compensation for this channel is 4.6 s, which is smaller than that of the ammonia channel. Taking the larger value, the upper limit of speed remains unchanged. Under the same 2m resolution requirement, the upper limit of speed for Comparative Example 3 is only 0.011 m / s, and the single-trip inspection time reaches 145.5 min. The above results show that the decomposition relationship of S13 does not contain undetermined coefficients. The spatial distortion directions of the outbound and return journeys are opposite but equal in magnitude. Their relative displacement is strictly equal to twice the product of the total equivalent time delay and the velocity. Therefore, the channel parameters can be solved by the reciprocating motion of the inspection mechanism itself without stopping the machine for ventilation. The inversion of S22 reduces the equivalent time constant. Based on this, S15 quantitatively converts the spatial resolution requirement into a velocity constraint. The stricter the resolution requirement, the lower the velocity limit. The quantitative relationship between the two can be verified column by column in Table 2.

[0091] Experimental Example 2: Comparability of Spatial Positioning Error and Profiles under Variable Speed ​​Conditions. This experimental example uses the methods of Example 1 and Comparative Examples 1 to 3. At the orbital coordinate of 47.0m, i.e., the center of segment 24, carbon dioxide was released at a constant flow rate using a calibrated thermal mass flow controller to form a concentration peak with a known location. Ammonia gas was released at the same coordinate via a permeation tube in the ammonia channel for verification. The reference concentration was determined by laboratory spectrophotometry. All four methods were run 10 times at the same speed of 0.090m / s. Subsequently, keeping the release conditions unchanged, a static reference profile was first obtained by stationing the inspection unit at each point, and then run 10 times each at three speeds: 0.027m / s, 0.054m / s, and 0.090m / s. The experimental results are shown in Tables 3 and 4.

[0092] Table 3, Spatial Positioning Error: Carbon dioxide channel reports peak position / m 47.13 49.11 47.28 49.38 CO2 channel positioning error / m 0.13 2.11 0.28 2.38 CO2 channel standard deviation / m 0.08 0.10 0.09 0.11 Ammonia channel positioning error / m 0.28 5.11 0.31 5.42 Ammonia channel standard deviation / m 0.12 0.15 0.13 0.16

[0093] Table 4. Profile comparability under variable speed conditions, carbon dioxide channel: Static reference 47.00 4.2 47.00 4.2 0.027 47.09 4.3 47.72 4.8 0.054 47.11 4.3 48.43 5.5 0.090 47.13 4.4 49.38 6.3 Peak position range of three levels / m 0.04 none 1.66 none

[0094] In Table 3, the reported peak position is the location of the maximum value obtained by parabolic fitting interpolation of the parameter field at three points. The positioning error is the absolute value of the difference between the reported peak position and the true coordinates of the release source, and the standard deviation is the dispersion of 10 repetitions. In Table 4, the method for determining the peak position is the same as in Table 3. The half-width at half-maximum (WHM) is the track length traversed by the parameter field from the peak value to half of the peak value. The range of the three peak positions is the difference between the maximum and minimum values ​​of the peak positions obtained from the three speeds. As shown in Table 3, the positioning error of the carbon dioxide channel in Comparative Example 1 is 2.11m, which is close to the product of the inspection speed and the first-order time constant of the channel; the positioning error of Comparative Example 2 is 0.28m, which is close to the product of the inspection speed and the sampling and transport delay of the channel; Comparative Example 3 has both errors, and the positioning error is 2.38m; after applying both compensations simultaneously in Example 1, the positioning error is reduced to 0.13m, which is consistent with the residual magnitude calculated from the standard deviation of the first-order time constant calibration. Due to the larger time constant, the positioning error of the ammonia channel in Comparative Example 3 reached 5.42m, while that in Example 1 was 0.28m. As shown in Table 4, the peak position and half-width at half-maximum (HWHM) of Comparative Example 3 increased linearly with velocity, and the peak position range of the three speeds was 1.66m, making the profiles obtained at different speeds incomparable. In Example 1, the peak position range of the three speeds was 0.04m, and the HWHM remained stable at around 4.3m, which was close to the static reference profile. The above results indicate that the sensor inertia causes waveform broadening and centroid shift, which must be eliminated in the time domain by the inversion of S22. The sampling pipeline causes an overall delay without changing the waveform, which must be eliminated in the spatial domain by the translation of S23. The two affect different error components, and the absence of either one will leave a residual of the corresponding magnitude. Furthermore, since the inversion formula of S22 only contains two time quantities, the time constant and the sampling period, and not the velocity, the velocity-related part is completely isolated to the translation term of S23 and calculated according to the point-by-point instantaneous velocity. The parameter field is thus decoupled from the inspection speed. This property is the premise for the establishment of the speed adjustment of S45 and the same-layer baseline of S31.

[0095] Experiment 3: Interlayer and full-cycle consistency of anomaly detection sensitivity. This experiment used the methods of Example 1 and Comparative Examples 3 and 4. A locally known increase in ammonia was created by closing the air inlet window of the 3rd layer in section 24. Ten measurements were taken at each of the three age groups: 14 days, 28 days, and 42 days. The ammonia reference concentration was determined by laboratory spectrophotometry. Additionally, at 28 days, 1, 2, 4, and 8 anomalous sections were sequentially created in the 3rd layer to investigate the influence of the proportion of anomalous sections within the same layer. The robust baselines of layers 1 to 4 in the same cycle were 8.5 ppm, 9.8 ppm, 11.2 ppm, and 12.6 ppm at 28 days, respectively, with an interlayer difference of 4.1 ppm. The experimental results are shown in Tables 5 and 6.

[0096] Table 5. Interlayer and full-cycle consistency of anomaly detection sensitivity, ammonia channel, segment 24, layer 3: 6.0 0.6 9.0 3.37 2.97 Not exceeding the standard 28 11.2 1.1 16.8 3.43 3.01 Exceeding the standard 42 17.2 1.6 25.4 3.46 3.03 Exceeding the standard throughout the event

[0097] Table 6. Impact of the number of abnormal segments in the same layer on the anomaly detection score, 28 days, layer 3: 3.43 yes 3.01 yes 2 3.42 yes 2.72 no 4 3.38 yes 2.31 no 8 3.32 yes 1.80 no

[0098] In Table 5, the robust baseline is the median of the parameter field values ​​for each segment of the layer. The absolute deviation of the median reflects the normal dispersion of the layer. The anomaly score is the dimensionless value calculated according to formula S31. The judgment of Comparative Example 3 is the conclusion obtained based on a fixed absolute threshold of 15 ppm. In Table 6, the anomaly judgment is given according to the anomaly threshold 3. As shown in Table 5, the anomaly scores of Example 1 are 3.37, 3.43, and 3.46 in the three age groups, with fluctuations not exceeding 0.09, while the robust baseline increases from 6.0 ppm to 17.2 ppm. Comparative Example 3 was missed at 14 days because the measured value of the anomaly segment was lower than the fixed threshold, and could not be located at 42 days because the baseline of the entire field exceeded the fixed threshold. Although the score of Comparative Example 4 is close to 3 when there is a single anomaly, it is affected by the anomaly sample itself. As shown in Table 6, when the number of anomalous segments in the same layer increased from 1 to 8, the score of Example 1 decreased from 3.43 to 3.32 but remained above the anomaly threshold. The score of Comparative Example 4 decreased from 3.01 to 1.80 and was missed when there were only 2 anomalous segments. The above results indicate that S31 uses the median and median absolute deviation of each segment in the same layer and round as the baseline. The inter-layer systematic differences and the baseline rise with age are automatically deducted, so the anomaly threshold does not need to be adjusted with layer number and age. Furthermore, since the breakdown point of the median and median absolute deviation is 50%, the estimator is still bounded when less than half of the segments are contaminated. A few strong anomalous segments will not raise the anomaly threshold, while the arithmetic mean and standard deviation will be pulled by the anomalous samples themselves. The more anomalous segments there are, the stronger the pull, and the score will decrease accordingly.

[0099] Experiment Example 4: Consistency Verification of Actuator Commands and Actual Actions, and Fault Source Location. This experiment example adopts the methods of Example 1, Comparative Examples 3 and 5. First, under fault-free conditions, images were repeatedly acquired 20 times at the same fixed opening degree, and the angle between the window panel and the window frame was measured. The standard deviation was taken as the repeatability of the image angle measurement. Then, the window was driven to the same commanded opening degree from both the opening and closing directions, and the difference in the angle was measured as the mechanical hysteresis. The two components were combined using the root of square to obtain the consistency threshold according to the common practice of measurement uncertainty assessment. Subsequently, a total of 12 known faults were set, including 6 instances where the air inlet window actuator linkage pin was pulled out and the window panel was fixed, 3 instances where the exhaust fan field isolation switch was disconnected but the programmable logic controller maintained the running command, and 3 instances where the air supply guide was deflected but the window operated normally. An additional 12 fault-free controls were also set. The experimental results are shown in Tables 7 and 8.

[0100] Table 7. Results of synthesizing actuator consistency thresholds and opening measurement: Small window full travel angle range / ° 8～62 Corresponding to 0 to 100 From fully closed to fully open Image angle measurement repeatability standard deviation / ° 0.4 0.7 20 repetitions After expansion with a inclusion factor of 2 / ° 0.8 1.5 Angle measurement components Mechanical hysteresis / ° 3.8 7.0 Difference between bidirectional drives Synthesis consistency threshold / percentage points 7.2 Value 8 Combination of square and root Actual aperture percentage measured at illuminance of 5 lx / % 31.2 Deviation 0.6 Reference value: 30.6 Actual aperture percentage measured at 20 lx illuminance / 30.8 Deviation 0.2 Reference value: 30.6 Actual aperture percentage measured at 200 lx illuminance / 30.5 Deviation −0.1 Reference value: 30.6 / % was measured after the lens had been running for 30 days without being cleaned 32.4 Deviation 1.8 Reference value: 30.6 Lens running for 60 days without cleaning Edge detection failure rate: 15% none This verification is invalid and a cleanup request is made.

[0101] Table 8. Fault detection time and fault source location results: Small window motion deviation detection time / min 22 186 Not detected Exhaust fan not rotating detection time / min 18 74 132 The number of correctly located sections is 12. 12 10 0 The number of correct fault types is 12 in total. 11 4 0 Track length requiring on-site inspection / m 2 2 96 Number of misjudgments in 12 fault-free comparisons 0 0 3

[0102] In Table 7, the converted percentage points are calculated based on a 100% conversion of 54° across the entire stroke. The reference value is obtained by converting the reading from the digital angle gauge. The composite consistency threshold is the root of the square of the angle measurement component and the mechanical hysteresis component, and the value is rounded up before being written to the controller. In Table 8, the detection time is the time elapsed from setting the fault to outputting a first-level warning. The number of correct segment locations is the number of times the output segment number matches the actual fault segment. The number of correct fault types is the number of times the output fault type matches the actual fault. The track length requiring on-site inspection is the length that needs to be covered by on-site inspection according to the output conclusion. Table 7 shows that the consistency threshold is 8 percentage points. After the fault setting, the commanded opening is 80%, and the actual opening measured by the image is about 31%, with a difference of about 49 percentage points, which is much greater than the threshold. Within the illumination range of 5 lx to 200 lx, the opening measurement deviation does not exceed 0.6 percentage points. When the lens has been running continuously for 30 days without cleaning, the deviation is 1.8 percentage points, which is still less than the threshold. When the lens has been running continuously for 60 days without cleaning, the edge detection failure rate reaches 15%. At this time, the current verification is deemed invalid and cleaning is requested. Table 8 shows that the detection time for small window movement deviation and exhaust fan non-rotation in Example 1 is 22 min and 18 min, respectively. In Comparative Example 5, the detection times are 186 min and 74 min, respectively. Comparative Example 3 failed to detect small window movement deviation. The number of correct segment positioning is 12 in Example 1, 10 in Comparative Example 5, and 0 in Comparative Example 3. The number of correct fault types is 11 in Example 1 and only 4 in Comparative Example 5. Comparative Example 3, due to the use of a fixed absolute threshold, misjudged 3 times in the later stage of feeding in 12 fault-free controls.

[0103] The above results indicate that the environmental control system operates in an open-loop manner for the air inlet window and exhaust fan. The controller only provides commands without mechanical position or speed feedback. S42 and S43 independently obtain the actual operating status visually, and S44 compares it with commands from the same control cycle. Faults are detected before environmental parameters accumulate to exceed limits, thus shortening the detection time by an order of magnitude. S33 relies on the distribution pattern of anomaly scores along the aquaculture layer and along the track. The air supply path synchronously points to the longitudinal position in the layer direction, while the equipment in the track direction synchronously points to the layer. This narrows the scope to a segment. Whether the actuator is not moving or the path is blocked needs to be distinguished by S44 using visually measured status and the stationary wind speed obtained in S41. The two provide location information and part information respectively, which are combined to give the fault source with a part number. In Example 1, one location was misjudged as an offset of the air supply guide, but it was actually caused by material accumulation in the material line blocking the air supply outlet, which is a form of obstruction not covered by the visual field.

[0104] As can be seen from Tables 1 to 8, the online identification of S1 provides S2 with channel parameters consistent with the step calibration. The dynamic compensation and coordinate reregistration of S2 reduce the positioning error from 2.38m to 0.13m and decouple the parameter field from the inspection speed. Based on this, S15 increases the upper limit of the cruise speed from 0.011m / s to 0.090m / s. The robustness score of S31 stabilizes the anomaly score at around 3.4 under the condition that the inter-layer baseline difference is 4.1ppm and the full-cycle baseline increases from 6.0ppm to 17.2ppm. S42 to S44 shorten the fault detection time from 186min to 22min and increase the number of correct fault types from 4 to 11.

[0105] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A track-based, end-to-end inspection and early warning system for broiler chicken integrated farming, characterized in that: include: The components include a track inspection mechanism, a layered environmental acquisition component, an equipment vision inspection component, an industrial communication interface, an early warning output unit, and an edge controller. The track inspection mechanism includes guide rails laid longitudinally along the top or side of the stacked breeding layers in the aquaculture workshop, inspection units that roll in cooperation with the guide rails via traveling wheels, a drive mechanism that drives the inspection units to operate at bidirectional speeds, an incremental encoder for measuring the displacement, position reference components set at equal intervals along the guide rails, and corresponding reading components located in the inspection units. The tiered environmental data acquisition system includes a lifting sensor pod suspended from the inspection unit via a lifting mechanism, with each stopping position corresponding to a sampling height of each aquaculture layer, as well as a gas sampling and detection component, a temperature and humidity sensor, and a wind speed sensor located within the pod. The equipment visual inspection component is located in the inspection unit and is used to independently acquire the actual operating status of the environmental control actuators. The industrial communication interface communicates with the programmable logic controller (PLC) of the environmental control system to obtain the instruction status and timestamps of the actuators. The edge controller serves as the edge control node in an industrial control system. Its inputs are connected to an incremental encoder, a reader, a hierarchical environmental acquisition component, a device vision inspection component, and an industrial communication interface. Its outputs are connected to the drive mechanism, the lifting mechanism, and the early warning output unit. It also includes a channel identification module for sequential data association, a dynamic compensation and spatial re-registration module, an anomaly diagnosis and gating module, an actuator consistency verification module, and an inspection scheduling module. Based on the abnormal diagnosis results and the verification results of the command status and the actual action status, the inspection and scheduling module adjusts the running speed and dwell position of the inspection unit and the stopping position of the lifting sensor pod, outputs the fault source and warning level to the early warning output unit, and sends back the operating mode flag of this round to the channel identification module, forming a closed loop of data collection, diagnosis and scheduling.

2. The inspection and early warning system according to claim 1, characterized in that, The position reference component is a passive magnet or an RFID tag, and the reader is a corresponding Hall switch or an RFID reader head. The drive mechanism includes a servo geared motor and a friction wheel drive assembly or a synchronous belt drive assembly connected to its output shaft. The traveling wheel rolls with the guide rail surface and bears the weight of the inspection unit. The friction wheel drive assembly or synchronous belt drive assembly is used to transmit traction force without bearing the weight of the inspection unit. The incremental encoder is coaxially connected to the output shaft of the servo geared motor, and its output end is connected to the dynamic compensation and spatial re-registration module along with the output end of the reader. The output signal of the reader is used to perform absolute position correction on the cumulative displacement of the incremental encoder.

3. The inspection and early warning system according to claim 1, characterized in that, The gas sampling and detection assembly includes a sampling port equipped with a dust filter, a sampling pipeline, a gas sensor, and a sampling pump. The gas sensor includes an electrochemical ammonia sensor and a non-dispersive infrared carbon dioxide sensor. The gas chambers of both are connected in parallel to the sampling pipeline via their respective flow restrictors, so that the two gas channels have their own independent branch volumes and branch flow rates. The sampling pump is connected to the downstream side of the two gas chambers and makes the sampling gas flow unidirectionally from the sampling port to the gas chambers. The temperature and humidity sensor is directly exposed to the ventilation window of the lifting sensing pod without passing through the sampling pipeline. The wind speed sensor is a hot-film wind speed sensor, and its output end is connected to the actuator consistency verification module.

4. The inspection and early warning system according to claim 1, characterized in that, The device's visual inspection component includes a visible light camera fixed to a two-degree-of-freedom gimbal, a supplementary light, and a dust cover. The two-degree-of-freedom gimbal is positioned to the side of the inspection unit and its rotation and pitch angles are controlled by the edge controller. The edge controller stores a calibration mapping between the orbital coordinates of each environmental control actuator and the corresponding gimbal angle. When the inspection unit is stationary, the optical axis of the visible light camera is aligned with the junction of the window panel and window frame of the air inlet window or with the air outlet of the exhaust fan according to this calibration mapping. The image output terminal of the visible light camera is connected to the actuator consistency verification module.

5. The inspection and early warning system according to claim 4, characterized in that, The industrial communication interface is an industrial Ethernet interface. The data it reads includes the opening command of the air inlet window, the start / stop and frequency commands of the exhaust fan, and their command timestamps. The data it sends back includes the fault source location number and the warning level. The three input terminals of the actuator consistency verification module are respectively connected to the equipment visual inspection component, the industrial communication interface, and the wind speed sensor. Its output terminals are respectively connected to the inspection scheduling module and the warning output unit. The warning output unit includes a field audible and visual alarm and a host computer monitoring terminal. The edge controller is also connected to a local storage unit, which stores the original measurement sequence, the compensated sequence, the calibration parameters, and the supplementary inspection queue for each channel.

6. A method for track-based full-process inspection and early warning in broiler three-dimensional farming, employing the system described in any one of claims 1-5, characterized in that, Includes the following steps: S1, During the period when the command status of the environmental control actuator remains unchanged, each track segment runs at the cruise speed and there are identifiable spatial changes in environmental parameters along the track, the control inspection unit completes adjacent outbound and return acquisitions at the same constant speed. For each environmental parameter channel, the uncompensated outbound and return profiles of the channel are normalized and cross-correlated. The total equivalent time delay of the channel is identified by the spatial displacement corresponding to the peak value of the correlation coefficient. After deducting the sampling and transport delay of the channel, its first-order time constant is obtained, and the cruise speed upper limit is set according to the equivalent time constant after compensation for each channel. S2, after a single-pass acquisition is completed, the sensor output of each channel is filtered and the first-order dynamic inversion is performed according to the first-order time constant of each channel to eliminate the measurement tail caused by sensor inertia. The track coordinates of each sampling point are corrected according to the instantaneous speed and the sampling and transport delay of the channel. The cumulative displacement error of the encoder is corrected with the position reference component. Then, the layer and section parameter fields of each channel are generated by resampling according to the breeding layer and track section. S3 calculates the anomaly score for each channel's parameter field based on the median and median absolute deviation of each section of the same breeding layer. After the command state is switched, the overall alarm is gated within the set time determined by the net volume of the breeding workshop and the effective exhaust volume. The suspected anomaly location and suspected anomaly type are output according to the distribution pattern of the anomaly score along the breeding layer direction and along the track direction. S4, the control inspection unit stays and retests at suspected abnormal locations and corresponding actuators. The equipment vision inspection component independently obtains the actual action status of the actuator, performs consistency verification with the command status of the same control cycle, and merges it with the result of S3. It outputs the fault source and warning level, and updates the next round of patrol, detailed inspection or stay operation mode and supplementary inspection queue for each track section according to the comprehensive value of the abnormality score of each channel.

7. The inspection and early warning method according to claim 6, characterized in that, The environmental parameter channels include an ammonia channel, a carbon dioxide channel, a temperature channel, and a humidity channel. Indicating the channel number, S1 includes: S11, when the command state remains unchanged within a time window consisting of a one-way trip and an adjacent return trip, when each track segment runs at cruising speed within the time window, when the duration of the time window is not greater than the quotient of the net volume of the aquaculture workshop and the current effective exhaust volume, and when the standard deviation of the outbound measurement sequence along the track coordinates is not less than three times the resolution of the channel sensor, a calibration permission signal is issued; when no time window meeting the above conditions appears within the set calibration period, a dedicated calibration round trip is arranged in which each track segment runs at cruising speed; S12, control the inspection unit to operate at a constant speed The outward journey is completed along the increasing direction of the orbital coordinates, and then the return journey is completed at the same constant speed. For the ... The original measurements of the channel without dynamic compensation are resampled at the same track coordinate intervals to obtain outbound and return profiles. For each candidate spatial displacement, the outbound profile is translated along the track coordinates by that displacement, and a normalized cross-correlation coefficient is calculated between the outbound profile and the return profile within their effective overlap interval. The value range of the candidate spatial displacement does not exceed one-quarter of the profile length, and the number of resampling points within the effective overlap interval is not less than a set lower limit for the number of overlap points. The spatial displacement corresponding to the peak value of this coefficient is taken. and that peak value ; S13, decomposed according to the following formulas, yields the... The first-order time constant of the channel sensor: ； In the formula, For the first The total equivalent delay of the channel; The first obtained for S12 Peak displacement of the channel; The constant inspection speed during calibration; For the first The first-order time constant of the channel sensor; For the first The sampling and transport delay of the channel is obtained by measuring the pure delay time when a standard gas of known concentration is introduced into the sampling port of the channel during the installation and commissioning phase to form a step input. It is corrected according to the ratio of the sampling branch flow rate at the calibration time to the current sampling branch flow rate. The temperature and humidity channels that do not pass through the sampling pipeline are taken as zero. S14, in Not less than the relevant threshold , Greater than zero and Greater than Time to determine the first The channel identification was valid this time, and the results obtained this time will be used to further improve the channel. The saved value is updated after being weighted and averaged with the channel's saved value at a ratio of 1:9; if it is determined to be invalid, the original saved value is retained and the channel's calibration invalidation flag is set; when the saved value is greater than twice the channel's calibration value during the installation and commissioning phase, a channel degradation warning is output; where, The first obtained for S12 Peak value of channel normalized cross-correlation coefficient; The relevant threshold is determined by the lower bound of the distribution of peak values ​​obtained from multiple consecutive round trips during the installation and commissioning phase under normal operating conditions. S15, set the maximum cruising speed using the following formula: ； In the formula, This is the maximum cruising speed. To achieve the required spatial resolution, the longitudinal length of a cage unit is taken. The response completion coefficient is set to 3, which corresponds to the first-order element completing 95.0% of the step response. For the first The equivalent time constant of the channel after S2 compensation is determined by applying a known step excitation to the channel during the installation and commissioning phase, taking the time it takes for the compensated output of the channel to change from the initial value to 63.2% of the final value, and then recalibrating it after replacing the channel sensor or dust filter. This indicates that the larger of the channel values ​​within the curly braces is selected.

8. The inspection and early warning method according to claim 7, characterized in that, In step S2, sampling and orbit coordinate correction are performed in real time during the movement of the inspection unit, while filtering, inversion, and reregistration are performed in batches on the sequence of a single journey after the single-journey acquisition is completed. Step S2 includes: S21, for the first The channel uses the sampling period of that channel. Synchronously record sensor output and orbital coordinates With instantaneous velocity When the reader detects a position reference, the track coordinates are set to the calibrated position of that reference, and the scaling factor is updated by the ratio of the distance between two adjacent position references to their corresponding encoder pulse increments, used for real-time conversion of track coordinates within the interval after that reference; the wind speed sensor does not participate in the acquisition during travel, and its measurement values ​​are acquired in the dwell state of S41; where, For the first The sampling period of a channel is the data output period of the sensor in that channel. The sampling sequence number; For the first The corrected orbital coordinates of each sampling point; For the first The instantaneous velocity of each sampling point; S22, after a single-trip data collection is completed, the first data collection point within that single trip... The channel sequence is first averaged once along the increasing sampling number direction and then once along the decreasing sampling number direction. The two ends of the sequence are extended according to the endpoint values. The window length of the moving average is the odd number of points corresponding to the quotient of one-tenth of the first-order time constant of the channel and the sampling period of the channel, rounded to the nearest integer, to obtain the filtered sequence. Then, a first-order dynamic inversion is performed according to the following formula, and the inversion result is limited to the range of the sensor of the channel: ； In the formula, For the first Channel 1 Compensation value for each sampling point; and The first Channel 1 The and the first The filtered value of each sampling point; The first obtained for S1 The channel's first-order time constant; For the first The sampling period of the channel; filtering once in both directions to make the net phase lag of the single program sequence zero; S23, assign each compensation value back to its actual sampling position according to the following formula, and then... Using the segment length as the unit and each stop position of the lifting sensor pod as the aquaculture layer number, the arithmetic mean of the compensation values ​​for pods falling into the same segment of the same aquaculture layer is taken to generate the first... Channel layer and section parameter fields : ； In the formula, For the first Channel 1 The orbital coordinates of each sampling point after reregistration; For the first The orbital coordinates of each sampling point are obtained from S21; The sign function is +1 when the inspection unit runs along the direction of increasing track coordinates and -1 when running along the direction of decreasing track coordinates; For the first The absolute value of the instantaneous velocity at each sampling point; For the first Sampling and transport delay of the channel; For the first The installation offset of the channel sampling port relative to the positioning reference point of the inspection unit is positive in the direction of increasing track coordinates; The segment length is the same as the spatial resolution required in S15. For the first Channel 1 Aquaculture layer The parameter field values ​​of the section; For breeding layer number, The numbers are segment numbers, all of which are dimensionless; the first-order dynamic inversion is performed in the time domain while the position translation is performed point by point according to the instantaneous velocity. The parameter fields of each channel obtained in this step are independent of the inspection speed.

9. The inspection and early warning method according to claim 8, characterized in that, S3 includes: S31, for the first For each culture layer in the channel, the median of the parameter field values ​​of each segment of that culture layer is taken as the robust baseline of that culture layer. The median absolute deviation of the culture layer is taken as the absolute value of the absolute difference between the parameter field values ​​of each section and the robust baseline. Then, calculate the anomaly score for each section according to the following formula, and take the larger of the absolute values ​​of the anomaly scores of each channel in the same section of the same breeding layer as the comprehensive score for that section of the breeding layer. : ； In the formula, For the first Channel 1 Aquaculture layer Abnormal scoring for the section; The parameter field values ​​obtained from S2; For the first Channel 1 A robust baseline for the culture layer; For the first Channel 1 The absolute deviation of the median in the culture layer; 0.6745 is the conversion constant for converting the absolute deviation of the median to the standard deviation, taken from the values ​​of the quartiles of the standard normal distribution; This indicates that the larger of the values ​​within the curly braces is selected. For the first Resolution of the channel sensor; For the first Aquaculture layer The comprehensive score of the section; the absolute value of the abnormal score is not less than the abnormality threshold. The passage, the aquaculture layer, and the corresponding section were determined to be abnormal. Take 3; S32, starting from the moment the instruction state changes, within the set time determined by the following formula. No overall alarm will be output within this period: ； In the formula, For setting time; To fine-tune the multiplier, during the installation and commissioning phase, a step increase in the number of operating units was applied to the exhaust fans. The time taken for each channel parameter to reach 95% of the new steady state was recorded, along with the time taken at that moment. and The ratio of the quotients is taken as the larger of the ratios obtained from each channel; This refers to the net volume of the breeding workshop; To determine the effective exhaust volume, the sum of the number of operating units of each type of exhaust fan and the rated air volume of that type of exhaust fan under the current working static pressure is taken. S33, for each channel, determine the abnormal score according to the spatial distribution pattern: the abnormal scores of each culture layer in the same segment on the same channel have the same sign and their absolute values ​​are not less than 1. When an abnormality is detected in a section, the section number is output and the section is considered a suspected abnormality. The absolute value of the abnormality score in at least three consecutive sections on the same channel within the same aquaculture layer must be no less than [amount missing]. The absolute values ​​of the anomaly scores for the remaining aquaculture layers in the same section were less than [a certain value]. When the culture layer is suspected of having an interlayer environmental anomaly, the culture layer number and section range are output; if the robust baseline of each culture layer on the same channel simultaneously exceeds the process setting range, and the maximum absolute value of the channel anomaly score is less than [missing information], then [missing information]. If the overall operating condition deviates, no output will be given within the validity period of the gate signal in S32; this step outputs the suspected abnormal location and suspected abnormal type, and the specific fault source is determined by S4.

10. The inspection and early warning method according to claim 9, characterized in that, S4 includes: S41, the overall score in a certain section is not less than When the value is twice that of the previous value, or when S33 determines that the section is suspected to be an abnormal air supply path or an abnormal inter-floor environment, the inspection unit is controlled to stay in the section. During the stay, the lifting sensor pod is controlled to stop at each floor, and the steady-state remeasurement value of each channel and the wind speed value at the stay position are collected. The stay time is the product of the larger of the response completion coefficient and the equivalent time constant after compensation of each channel. S42, during the installation and commissioning phase, at least five air inlet window opening commands, evenly distributed from fully closed to fully open, are issued sequentially. After each command is executed, the visible light camera acquires an image and measures the angle between the window panel and the window frame, establishing a calibration mapping between the command opening and the window panel angle. During operation, edge detection is performed on the image of the air inlet window at the stationary position, the straight line segment between the window panel and the window frame is extracted, and its angle is measured. The actual opening is obtained by piecewise linear interpolation on the calibration mapping. , The unit is percentage; S43, acquire image sequences of the exhaust fan outlet area at a set frame rate, and calculate normalized inter-frame motion features at two non-integer multiple frame intervals respectively; during the installation and commissioning phase, measure the stop feature value and the operation feature value at the same two frame intervals in both the stop and operation states of the exhaust fan; when the operation feature value at the same frame interval is greater than the stop feature value, and the motion features measured at the operation in both frame intervals are less than the arithmetic mean of the stop feature value and the operation feature value corresponding to that frame interval, it is determined that the exhaust fan blades are not rotating; S44, Obtain the opening degree of the air intake window command within the same control cycle as the image acquisition time. In conjunction with the start / stop commands of the exhaust fan, and The absolute value of the difference is greater than the consistency threshold. If, at any time, or when the exhaust fan start / stop command is "run" but S43 determines that the blades are not rotating, it is determined that the actuator has a deviation fault where the command and actual action are inconsistent. In the formula, The opening degree of the air intake window command issued by the programmable logic controller; The consistency threshold is determined by combining the measured mechanical backlash of the air inlet window actuator linkage mechanism with the repeatability measured value of the image angle measurement in S42 using the square root. Then, the output is fused according to the following correspondence: when a suspected abnormal air supply path is determined and the air inlet window in that section is determined to have a deviation fault, output a fault in the air inlet window actuator of that section; when a suspected abnormal air supply path is determined, the air inlet window in that section is not determined to have a deviation fault, and the wind speed at the residence position obtained in S41 is lower than the median wind speed at each residence position in that aquaculture layer, output a blocked ventilation path in that section; when an overall operating condition is determined to be deviated and there is an exhaust fan whose start / stop command is "run" but whose blades are not rotating, output a transmission or electrical fault in that exhaust fan. S45, based on the comprehensive score, sets the operating mode for each section in the next round of inspections: comprehensive score less than The section operates at the maximum cruising speed, not less than Sections less than twice the speed of the cruise speed are inspected at a speed determined by the ratio of the cruise speed limit and the response completion coefficient, and image acquisition is encrypted. Sections not less than twice the speed of the cruise speed are kept in S41. When the scheduling time of a single round of inspection exceeds the set value, the sections to be handled in this round are selected from the highest to the lowest comprehensive score, and the sequence number of the delayed section is written into the supplementary inspection queue. The sections in the supplementary inspection queue are arranged in the next round of inspection with priority over sections with the same score and are re-inspected within no more than two rounds of inspection. When the same fault source recurs in at least two consecutive rounds of inspection, a level 2 warning is output. When an action deviation fault is determined in S44, a level 1 warning with priority over the level 2 warning is output.

Images