Flow estimation of flowing solids in a substantially horizontal pipe

Abstract

An apparatus is provided for estimating the flow rate of a solid substance flowing in a generally horizontal pipe having a movable conveying element, such as an auger in the pipe, for conveying the substance along the pipe. The apparatus includes a distance sensor mounted on the upper part of the pipe and arranged to detect the distance between the substance below the pipe and the sensor. Another sensor is provided to detect the moving speed of the conveying element by, for example, detecting the metal of the auger blades. The output from the sensor is fed to a control system for analyzing the output signal from the sensor over time to provide a flow rate estimate. The sensor can be mounted in a pressurized housing to prevent dust from escaping through the sensor's openings.

Description

Technical Field

[0001] This application relates to a method for indirectly measuring the flow rate of a liquid substance in a non-vertical pipe. Background Technology

[0002] To date, feed remains the most significant expense in intensive livestock production, typically accounting for 60% to 70% of total production costs. However, in modern commercial livestock production, feed remains one of the most unpredictable aspects of the entire process. This significantly reduces efficiency across various areas of production. For example, the lack of such data on feed inventory management often presents challenges such as feed disruptions, high costs associated with emergency feed deliveries, the discovery of unused feed during inventory clearance, inconsistent feed delivery schedules, and other challenges. The availability of highly accurate, near-real-time feed usage data, rather than the current practice of using data from months prior that is only available after the growth cycle is complete, is crucial for proactively adjusting feeding programs, environmental factors, and other key production inputs. Other areas that will directly benefit from the high availability of highly accurate, near-real-time feed usage data include enhanced animal performance analysis and planning, faster detection of animal health problems, and the application of micronutrients (such as enzymes, probiotics, antioxidants, etc.) on farms rather than in feed mills, among many others.

[0003] Current methods for monitoring feed consumption primarily rely on experience, historical data, and approximations using information from historical feed orders and knowledge of animal inventory in a given barn. Another common method is to monitor feed loss in the barn using expensive weighing sensors.

[0004] Therefore, there is a need for a low-cost solution that can measure the flow rate of flowing solids with high accuracy in near real-time, especially in environments commonly found in commercial livestock farming. Summary of the Invention

[0005] According to the present invention, an apparatus is provided for estimating the flow rate of a solid substance flowing in a generally horizontal pipe, the pipe having a movable conveying element for conveying the substance along the pipe, the apparatus comprising:

[0006] Multiple distance sensors are installed on the upper part of the pipe to detect the distance between the material below the pipe and the sensors;

[0007] A control system is used to analyze the output signals from the sensors over time to provide an estimate of the flow rate.

[0008] The arrangement disclosed herein uses a range detection sensor to determine the flow rate of a flowing solid substance in a pipe. The sensor, which detects the distance between the substance and the sensor, is placed at the top of the pipe and points towards the inner bottom. When a movable element (e.g., an auger or disc) below the sensor is detected, a value is read from the range detection sensor to estimate the velocity of the flowing solid substance within the pipe; otherwise, the reading is extrapolated from the height of the substance in the pipe.

[0009] Optionally, a sensor device is provided that detects the movement speed of a feeding system using a moving element in the pipe. This can be done with greater precision using specific or discrete sensors that react to the moving element. This can preferably be achieved by using a metal detection sensor for a metal auger. However, other systems can use rotation / motion sensors, accelerometers, etc., connected to the engine shaft. Using a sensor array can provide more accurate readings, reduce the signal-to-noise ratio (SNR), and display outliers. The sensors are driven by a microcontroller. Several algorithms for the data from the sensors output the amount of flowing solid material through the pipe at fixed time intervals.

[0010] The method / process for acquiring data and determining the flow rate of solids in a pipeline includes the following steps:

[0011] --The distance / range sensor continuously collects data.

[0012] --Optional dedicated sensors detect the speed of the conveying elements.

[0013] --Data is extracted from the sensor. The extracted data is labeled as raw data and has been preprocessed using a specific data extraction algorithm.

[0014] --Calibration data is used for extraction.

[0015] --Place the raw data into the storage buffer.

[0016] --The data is processed using a flow determination algorithm stored in the buffer.

[0017] --Configuration data is used to refine the results of the traffic algorithm.

[0018] The independent yet combined traffic prediction methods can be applied as follows:

[0019] --It can measure flowing solids, where flowing matter is described as dry solid granular matter or powder that can flow. Examples of such flowing matter may include, but are not limited to, animal feed, grains, seeds, plastic pellets, dry cement, flour, etc.

[0020] The arrangement described in this paper was developed for a specific application: measuring feed flow rates in spherical and pasty forms within the piping of an automated livestock feeding system. However, the system described herein can be used to measure other substances, and particle size will not affect the applicability of the arrangement described herein. If the arrangement is applied to very small particles, some adjustments may be required, particularly to the machine learning (ML) algorithm and possibly to the dustproof components of the arrangement described herein.

[0021] The flow rate of a flowing solid substance can be expressed in liters per second. -1 (Liters.s -1 (or kilograms per second) -1 (Kg.s -1 The processor is configured to calculate the weight and / or volume of a flowing solid substance based on its velocity and volume within the pipe. The volume of the flowing substance is calculated in liters and can be converted to weight (kilograms [kg]) using the substance's density. The processor can also adjust for any variations in the measured temperature / humidity of the flowing substance from standardized or reference values ​​(e.g., by adjusting the fluid's density and viscosity). The mass and weight of the measured substance are used interchangeably in the documentation and are associated with density.

[0022] The arrangement described herein can be used in any non-vertical closed pipe where the flowing material moves by means other than gravity and is not suitable for free-falling materials in vertical applications.

[0023] The arrangement described herein can be applied to situations where animal feed is moved by a conveyor belt, but is not limited to typical conveyor belts used in automated animal feeding systems commonly used in commercial livestock production. Such feeding systems typically have a conveying element that can move within a feed conduit for conveying feed, and multiple drop feeders spaced at intervals along the path of the feed conduit. Typically, the feed conduit is a closed, opaque, flexible feed tube (e.g., made of polyvinyl chloride (PVC)). The movable conveying element housed within the feed conduit is typically a centerless auger, which can be rotary-driven or axially driven, conveying feed through the tube in one direction via a motor with an output shaft.

[0024] This system is applicable to pipes and movable mechanical components made of any material, and is also suitable for applications where fluid substances are moved by pressure. Some adjustments to the software algorithm are required, particularly for the distance detection sensor, to accommodate these changes; however, the overall concept remains applicable.

[0025] The alternative and preferred embodiments further described herein refer to the application of feed flow measurement in the piping of an automated animal feeding system.

[0026] The sensor is mounted directly on the feeding system piping, preferably within the grain silo housing as further suggested in the instructions, and particularly on the feed piping at any desired location. No components are placed inside the feed piping.

[0027] Based on distance detection sensors, the currently preferred embodiment of the arrangement described herein typically requires a hole in the pipe for each sensor; however, it is possible to omit these holes. Currently, the diameter of the holes can be approximately 1 inch, but the diameter can be varied.

[0028] The sensor is located at the top of the tube, facing downwards towards the flowing solid material moving within the tube. As described in the sensor specification, each housing containing a distance detection sensor is spaced at a minimum distance of 100 mm from another such housing to avoid crosstalk between their enclosed sensors.

[0029] In the second arrangement, detection can be based on a vibration sensor instead of a distance sensor. The vibration sensor can be mounted on the outside of the pipe, thus eliminating the need for any holes. The sensor contacts the pipe, preferably within a housing, as described further below. Attached Figure Description

[0030] Embodiments of the present invention will now be described with reference to the accompanying drawings, wherein:

[0031] Figure 1 This is a schematic diagram outlining a first embodiment of the method according to the present invention.

[0032] Figure 2 yes Figure 1 A schematic diagram of the data processing method.

[0033] Figure 3 yes Figure 1 A schematic diagram outlining the flow prediction process of the method.

[0034] Figure 4 yes Figure 1 A schematic diagram of the distance detection acquisition and extraction scheme of the method.

[0035] Figure 5 yes Figure 1 A schematic diagram of the flow algorithm pipeline for the method.

[0036] Figure 6 yes Figure 5 A schematic diagram of the rotation memory buffer update process.

[0037] Figure 7 yes Figure 5 A schematic diagram outlining the algorithm configuration process of the method.

[0038] Figure 8 yes Figure 5A schematic diagram illustrating the data transformation process from data acquisition to raw data.

[0039] Figure 9 It is used for Figure 5 Isometric plot of the calibration box of the method.

[0040] Figure 10 yes Figure 1 A schematic diagram of the ideal distribution of the original data for the method.

[0041] Figure 11 Is with Figure 10 A schematic diagram of an example of an uncalibrated profile compared to the ideal profile of the original data.

[0042] Figure 12 yes Figure 1 A schematic diagram of the housing of the distance detection sensor for the method.

[0043] Figure 13 yes Figure 12 A schematic diagram of a first embodiment in which the sensor is preferably placed inside the housing.

[0044] Figure 14 yes Figure 12 A schematic diagram of a second embodiment in which the sensor is preferably placed inside the housing.

[0045] Figure 15 This is a schematic diagram outlining a second embodiment of the method according to the present invention.

[0046] Figure 16 yes Figure 15 A schematic diagram outlining the overall data flow of the method.

[0047] Figure 17 yes Figure 15 A schematic diagram of the flow estimation process using this method.

[0048] Figure 18 yes Figure 15 A schematic diagram of the transducer data extraction method.

[0049] Figure 19 yes Figure 1 and 15 A flowchart outlining the three-stage algorithm for the combination method.

[0050] Figure 20 yes Figure 20 An overview of the scoring phase of the algorithm.

[0051] Figure 21 This is an overview of the hybrid phase of graph algorithms.

[0052] Figure 22 This is an overview of the voting phase of graph algorithms.

[0053] In the accompanying drawings, the same reference numerals denote corresponding parts in different drawings. Detailed Implementation

[0054] like Figure 1 As shown, the disclosed arrangement described herein consists of the following components.

[0055] Example 10 includes a commercial distance sensor 12, preferably a short-range light-based sensor that uses a time-of-flight method to obtain accurate measurements. This could be a commercial lidar sensor of this type. Other distance detection sensors, such as ultrasonic sensors, could be used, but their accuracy may be significantly reduced due to their technological limitations. The sensor is mounted on a conduit 181 containing a flowable solid material 20, typically granular, such as animal feed, fed through the conduit by a conveying member, such as a auger blade 24. The conduit is typically horizontal, i.e., close enough to be horizontal that the material slides onto the bottom surface 181 of the conduit rather than remaining on the blade.

[0056] Preferably, the sensor array is provided with a set of three sensors 12, but the number can vary.

[0057] Data processing module 22 can use commercial host computer boards, such as those built around the ESP8266.

[0058] The array is mounted in a housing 14, which is characterized by having a dustproof device, such as a commercial blower used as described below.

[0059] Other alternative implementations may include using a dustproof glass plate, mounting a distance range detection sensor on a vibrating plate, etc. A lower power consumption and smaller size solution can be used instead of the blower.

[0060] The communication module 16 includes electronic components that allow communication with other computers using well-known communication protocols, including but not limited to WiFi, Bluetooth, Zigbee, and Ethernet.

[0061] The optional implementation of data processing used by the communication module is as follows: Figure 2 As shown.

[0062] External devices, with or without displays, can be configured to report / use the system's output. Examples include computers or external systems 18 that control feed dispensing, feed conveying systems, etc.

[0063] External storage devices can be used as external peripherals or any commonly used commercial online storage system.

[0064] Communication can be provided through standard industrial communication equipment and protocols.

[0065] Additional sensors can be added to monitor environmental parameters, including ambient humidity, temperature, pressure, and photocurrent. Data from these sensors is used to improve or refine feed flow prediction or readings from the range sensor and / or audio sensor array. Optionally, these sensors can be added to the main housing or another housing and communicate wirelessly. Data collected from these additional sensors is processed in parallel with the raw range and / or audio data in the feed flow prediction process described below for flow estimation.

[0066] Equipment 26 can be provided to assess feed density, shape, and overall quality.

[0067] Sensors can be configured to measure the composition of a substance (i.e., density, flowability, viscosity, and viscous properties) to improve and / or automate sensor configuration and enhance flow prediction accuracy.

[0068] The data processing module 22 on the custom hardware device (main computer board or MCB) can use specially manufactured processing units (CPU, MCU, etc.) for the application, such as field-programmable gate array (FPGA) chips, specifically programmed for the devices described herein, and / or ARM4xx or related MCUs, which are typically used in thin devices. This implementation will result in lower power consumption, higher CPU efficiency, and lower overall cost.

[0069] The method described in this paper includes several steps as described below. Steps 1a and 1b (optional), and 2a and 2b (optional) are performed in parallel.

[0070] Step 1a: Data acquisition from distance detection sensor

[0071] The distance detection sensor 12 measures the distance between a combination of optical components and a target in a specific direction. By mounting the sensor on a sensor housing, distance detection data can be acquired.

[0072] Preferably, the distance detection sensor is calibrated to obtain the most accurate data with the highest signal-to-noise ratio (SNR) for a specific application of the distance sensor housing. The calibration process is described below.

[0073] Step 1b: Metal Detection Data Collection

[0074] This optional step provides another sensor 28 that detects the movement of the auger within the feed pipe 181. The obtained data represents the distance between the metal detection sensor and the auger blades 24. Over time, as the auger 24 rotates, the signal changes and can be described as a waveform.

[0075] Step 2a: Distance detection data extraction

[0076] The distance sensors connect to the MCB using any communication protocol. The acquisition refresh rate is set to obtain sufficient data to filter out noise or any other artifacts caused by the sensor target or the sensor itself. A preferred acquisition refresh rate is 30 reads per second for each distance sensor. Optionally, each sensor is synchronized by an external clock to have the same refresh rate and timestamp.

[0077] The MCB is a very low-profile component. It has very limited memory, and each new value from the sensor replaces the previous value using a double-buffering method. The MCB performs a specific algorithm on the data acquired by the sensor and then stores it in memory for use in subsequent steps. This data is called the raw data. The algorithm will be described below.

[0078] MCB has a limited memory, so the amount of raw data stored is defined according to this limit. The allocated memory is shared between the processes in steps 1 and 2.

[0079] Step 2b: Extraction of linear auger speed

[0080] This optional step uses an algorithm capable of utilizing waveform signals acquired from the metal detection sensor 28. This algorithm converts the waveform into a linear velocity, representing the speed at which the auger 24 applies feed into the pipe. Furthermore, this linear velocity can be used to segment the feed system between when it is open, when the auger is rotating, or when it is closed and not rotating.

[0081] After this process, only the linear velocity is maintained and appended to the data extracted in step 2a to minimize the amount of MCB memory used.

[0082] Step 3: Calibration Data

[0083] Calibration data is used to extract distance data from the values ​​acquired by the distance detection sensor. The method for obtaining this data will be explained below in conjunction with calibration.

[0084] Step 4: Raw Data

[0085] The raw data is published by the extraction and acquisition process described in steps 1 and 2. This data represents the distance sensed by the distance detection sensor, indexed by a timestamp.

[0086] Because the MCB has a very limited memory capacity, this data is organized into a rotating buffer. Specific designs for memory management and optimization are implemented in the arrangement described herein. If unused data is replaced by new data, the memory manager can inform the process described in step 5. The memory manager can change the priority of processes accessing the original data (i.e., steps 2 and 5) to balance latency stability between data flushing (step 2) and feed stream algorithm data consumption (step 5). The memory manager also uses feedback from step 5 to free up memory by deleting recently processed data.

[0087] To address the MCB memory limit, this memory segment can also contain intermediate values ​​processed in step 5. Although this data has been technically altered, it is still referred to as the original data.

[0088] Step 5: Traffic Algorithm

[0089] Flow Algorithm (FRA) consists of several algorithms organized in a greedy pipeline. In this greedy pipeline, each deeper step does not have to wait until it has finished processing its block of data. Once data is marked as processed, regardless of the state of the entire data block, deeper steps can begin processing the data used by the steps that ran at lower depths.

[0090] FRA can manipulate the memory of the MCB allocated by the memory manager. The FRA process ensures that the non-primitive data it creates replaces the original data that has already been used. This non-primitive data consists of temporary values ​​used by the FRA pipeline. FRA guarantees that the size of the temporary value will not exceed the size of the data consumed to generate it. When the size of the temporary value is less than the size of the consumed data, the memory manager knows that memory is available.

[0091] Each distance detection sensor requires an FRA pipeline (optionally, each distance and velocity detection sensor pair requires an FRA pipeline). The output value of each pipeline is passed to a global Kalman filter-like process. The Kalman filter is used to reduce the uncertainty of the processed values. The output value of each FRA pipeline is used as a ground truth observation of the output value of another FRA pipeline.

[0092] An overview of FRA is as follows: Figure 5 As shown, an example using feed flow rate is presented.

[0093] Each method indicated in parentheses is a preferred implementation of the arrangement described herein. Alternative implementations may use different methods and / or skip and / or swap certain steps in the FRA. Such modifications may result in a faster FRA process, but some may lead to a decrease in accuracy. The greedy pipeline advances over the rotating memory buffer as follows: Figure 6As shown. Note that each step in the pipeline may modify the original data and free up some memory.

[0094] Step 6: Traffic Forecasting

[0095] The flow forecast calculated in step 5 is timestamped. This value is placed in a double-buffered memory. It can be retrieved by an external component, as specified in the optional steps below. The value is expressed in Kg.seconds. -1 Or Liters.second -1 This indicates feed flow rate. The output value is valid between its appended timestamp and the timestamp of the next output value.

[0096] Optional steps

[0097] Optionally, in an external reporting / monitoring system, the calculated feed flow value (as defined in step 6) can be further transmitted to an external reporting or monitoring system 18, which may or may not have a visual display.

[0098] external memory

[0099] Optionally, the arrangement described herein may provide output feed flow values ​​(as defined in step 6) for storage and future use. The storage may be a local device setup for the arrangement described herein or any conventional commercial online storage system.

[0100] Optionally, to improve the accuracy of flow calculation, the FRA can be configured on-site before the system starts, with the initial flow measurement run serving as the first run. This first configuration run will determine the parameters that the FRA will integrate into the flow calculation. Figure 7 This document provides an overview of the configuration process performed in the field during the initial operation of the layout described herein.

[0101] Figure 7 Steps 1 through 6 are as described above in the flow estimation process. Steps 7 through 9 are based on a process called the autoencoder process. In particular, step 7 tracks changes in the truth value and output value of the flow algorithm. This process maps the truth value to flow values ​​that encode the mapping as a function. The value in step 8 represents the truth value of the flow. This value can be derived from a nominal measure of the weight of the flowing solid material passing through the pipe, providing the total known weight or volume of the flowing solid material that has passed through the pipe.

[0102] There is no limit to the number of truth values ​​that can be used to configure the system. If the truth values ​​are expressed in volume, the density of the flowing solid material, for which the flow rate has been measured, needs to be input into the system to convert the volume to weight.

[0103] The mapped parameters encoded in the function are returned to the FRA. The system can then determine when it has enough data to ensure the encoded values ​​are sufficiently accurate to end the configuration process.

[0104] An algorithm is provided for extracting data from a distance sensing sensor and applying it to steps 1 and 2 of the flow estimation process described herein. The input data is referred to as distance data (DD), and the output data is referred to as raw data (RD). The distance range sensor data extraction algorithm continuously calculates RD data from the DD using calibration data, as further described regarding calibration. Figure 8 This explains how the data is transformed from collected data into raw data. This corrected data will be the raw data used in step 5 of the traffic forecasting process.

[0105] FRA is used in step 5 of the traffic forecasting process. The current implementation of the FRA pipeline has a depth of six steps. Each step modifies the raw data for the next step. Each step guarantees that the amount of memory used will not increase in any process. FRA considers timestamped blocks of data, and each step processes the data in the order of its timestamps. Each individual data block within each block is considered independent, even for FRA steps that use the entire data block. This allows each step to mark data as ready for the next deeper step, even if the current step is still using the data block. The goal is to keep the active pipeline as short as possible to obtain as much free memory as possible for new raw data.

[0106] refer to Figure 5 The first denoising filter I serves as the first denoising channel for the raw data, eliminating negligible mid-to-high frequency noise and artifacts, and is also used to interpolate the data over time to improve the resolution of the raw data. Optionally, if metal detection sensor data extraction is implemented as step Ib, this step can be renamed Ia.

[0107] Optionally, metal detection sensor data extraction Ib can be used in parallel with step Ia. Step Ib uses data from the metal detection sensor to determine the linear speed of the auger.

[0108] The envelope-like filter eliminates the sinusoidal artifact caused by the auger passing through pipe IIa. While eliminating this artifact, the auger's speed is also determined as IIb. Optionally, the implementation may combine the first denoising filter I and the envelope-like filter IIa.

[0109] Alternatively, step IIb can use data from a dedicated sensor to determine the auger speed; in this case, step IIb is not required and is replaced by step Ib. It can run at any time before step IV.

[0110] The second denoising filter III removes high-frequency noise that the filter may generate on the envelope signal.

[0111] Considering that the velocity of the flowing solid material through the pipe is the direct derivative of the auger velocity, velocity compensation IV of the flowing solid material is required. This filter uses the output from IIb, and if step IIb is not available, the output of Ib can be used to compensate for the value of the output in step III.

[0112] Flow segmentation V divides the data into two states: flowing or non-flowing solid matter. This method uses the output values ​​of both steps (IV) and (IIb, optional Ib). Flowing solid matter with a non-flowing state prevents the system from treating background noise as a minimum constant flow of solid matter. In this case, the flow rate is set to 0.

[0113] Volume Pipe Mapping (VI) uses 3D modeling of the feed flowing through the pipe to convert 2D data into a timestamped output from V, i.e., height to size in mm. 3 The volume is the volume between two timestamped 2D values ​​located below the distance detection sensor, as shown in step V. This volume is then converted into flow rate in mm using the velocity shown in step IIa, and optionally the velocity shown in IIb. 3 .s -1 This indicates that, if available, applying the density and viscosity of the flowing solid to the flow rate will give a result in kg.s. -1 .

[0114] The Kalman filter is a standard filter used to reduce the uncertainty of FRA calculations that use previous values ​​and other FRA pipeline values ​​as output.

[0115] Preferably, each distance detection sensor is calibrated to correct for its inherent error, which may lead to incorrect representation of distance within certain ranges. Specifically, in order to recreate accurate results, each distance range sensor used in this embodiment needs to be calibrated to match an arbitrary ideal profile.

[0116] Figure 9 The calibration cassette consists of a feed tube section 32, preferably 3 inches in diameter, with a 25mm wide slit 33 cut into the top of the tube. A sample of the auger / blade and feed 20 is placed inside the cassette. The cassette should have sections suitable for various feed levels, regardless of the amount of feed placed inside. The auger or feed does not move within the cassette. Preferably, the tube, auger, and feed used for the calibration cassette should be as close as possible to those used in the operating environment where the measurements will be performed.

[0117] Figure 9The calibration box depicted is one possible implementation for system testing. Various implementations are possible, with the goal of correcting sensor defects when aiming at various heights of augers, pipes, and flowing solid materials. Optionally, the calibration process can be run multiple times for the same sensor, with various types of feed placed in the calibration box at once, to improve the accuracy of distance sensor data acquisition.

[0118] By using different feed heights in the calibration chamber, an ideal profile can be created for the raw data. This profile represents the distance traveled between the distance range sensor and the material within the calibration chamber along the axis of the distance range sensor.

[0119] Figure 10 Showing with Figure 9 The example shown is an optional calibration box that achieves an ideal profile that matches the side view. An uncalibrated distance range sensor is placed on top of the opening of the calibration box. The sensor slides from one side to the other via a motor to ensure smooth travel. Data is captured from the sensor during this sliding motion. The sensor's linear velocity should match the velocity of the flowing solid material being measured through the pipe. A calibration conversion is calculated using the difference between the known profile and the measured uncalibrated profile. Feeding the uncalibrated data into this conversion returns data that looks more like the ideal data, such as... Figure 11 As shown.

[0120] The calibration kit is also used to find the correct configuration for the distance sensor, including refresh rate, threshold, etc., thereby improving the stability and accuracy of each sensor in the distance detection sensor housing.

[0121] A preferred embodiment of the distance sensor housing integrates three distance sensors that target a flowing solid material moving in the pipe at different angles and / or distances, as further described in this section. Alternative housing designs allow for more or fewer than three distance sensors. These all target feed or moving mechanical components in the pipe from different angles and / or distances and / or positions. Optionally, this embodiment may include one or more conveyor belt speed sensors.

[0122] An embodiment of the housing, for example Figure 12 As shown, it includes a blower 36, which is configured to create an overpressure zone within the housing 14. The overpressure is discharged into the duct through distance detection sensor array holes 38 and 40. The resulting airflow does not affect the flow of matter in the duct and prevents dust generated by the material flow from entering the housing.

[0123] Alternatively, the sensor array may include one or more speed detection sensors 28 for determining the linear velocity of the moving mechanical element.

[0124] To create the pressurized zone, the housing 14 is made of any airtight material. The housing is made as small as possible around its components to create the pressurized zone as quickly and stably as possible.

[0125] Because the distance detection sensor requires its beam to return to the sensor, holes need to be drilled in the pipe. To prevent feed from falling out of the pipe, holes are drilled at the top of the feed pipe. Holes 38 and 40 are preferably small enough to be covered if the range detection housing needs to be removed.

[0126] Figure 13 Figures 121, 122, and 123 illustrate the preferred placement of the three distance detection sensors 12 within the housing. Preferably, when distance detection sensors 122 and 123 are not vertically aligned, the angle of the analysis beam 124 of the distance detection sensors 122 and 123 shown is between 30° and 45°. As shown, the minimum spacing between sensors 121 and 123 is 10 mm. The distance between 121 and 122 is determined by the tube section between them.

[0127] Each sensor has multiple angles and positions, which is beneficial to the accuracy of the Kalman filter used in the flow detection algorithm as described above.

[0128] Alternative placement can change the angle and / or position of one or more sensors, such as Figure 14 As shown, or adjust to a larger or smaller number of distance detection sensors or any type of environmental sensor.

[0129] like Figure 15 As disclosed herein, additional information for use in this invention can be obtained, and the additional information consists of the following components.

[0130] A vibration sensor 42 is provided, which may include any vibration detector that captures frequencies from 1 Hz to 20 kHz, particularly, but not limited to, commercial audio pickups that record vibrations of solid objects rather than air. A preferred option is an audio pickup for music recording, as they are constructed to respond to the specific frequency band required for this embodiment. Alternatively, a commercial piezoelectric transducer may be used.

[0131] This embodiment requires only one sensor. Optionally, more sensors can be used to improve the reliability of the predicted traffic flow. A commercial amplifier 44 is used to limit signal distortion while allowing the analog-to-digital converter (ADC) to read the signal. The commercial ADC is used to convert the analog signal into a digital signal for audio preprocessing.

[0132] The data processing module consists of commercial processing units (MCUs, CPUs, etc.), similar to those used in smartphones or Internet of Things (IoT) devices, with pre-loaded data and machine learning (ML) algorithms. Preferably, the processing unit embeds the required ADC. The custom software algorithm described below is the pre-loaded ML data. The pre-loaded data is constructed using learning processes known in existing (SotA) machine learning techniques.

[0133] The housing has ample space to accommodate all components and includes a bracket for mechanically coupling the vibration sensor to the pipe without modifying the detected signal. No components are placed inside the pipe of the feed system itself. This allows for complete dust protection of the housing arrangement described herein without interfering with data acquisition, using appropriate dustproof insulation.

[0134] The connectivity / communication module is provided by electronic components that allow communication with other computers using standard communication protocols. This embodiment can use vibration sensors to determine the flow rate of flowing solids by applying machine learning (ML) methods to measure vibrations in the pipe (see [link]). Figure 16 ML algorithms must be trained using a large amount of data (data tensors and real data). During training, the ML service iteratively builds a predictive model. At the end of the training phase, the final published predictive model will output the value with the smallest error compared to the real data on its training data.

[0135] After training, the model will accurately predict flow rates based on new input data. For accuracy, the prediction model must be trained with relevant data. For correlation, the data must be correlated with the real data it is linked to. Component frequencies generated by the feed system are selected as the most relevant data to build an accurate prediction model. These vibrations can be acquired using an audio pickup placed on the feed pipe.

[0136] During flow estimation, a vibration transducer array continuously acquires data. Data is extracted from the vibration sensors to form a raw audio file of a usable standard waveform, which is then stored in the computer's memory. Audio preprocessing is performed by custom software to extract useful features for ML forward propagation. The audio features are stored in the computer's memory. ML forward propagation is then performed on the audio features.

[0137] Optionally, this process can be executed locally. Alternatively, ML forward propagation can be handled online. When using online ML forward propagation, the ML algorithm can be an updated version of itself based on the evolving SotA learning technique. This implementation requires subscribing to and connecting to an online ML service. ML preloading of the model is used for... Figure 17 Step 5. The traffic prediction is stored in the computer's memory.

[0138] Optionally, an external monitoring system can be used to report calculated feed flow rates. Optionally, feed flow rate forecasts can be stored locally or published online.

[0139] In machine learning methods, an implementation of a method for training a machine learning model is described. The ML model is trained on labeled data collected through the following process. The labeled data consists of processed audio data segments within a given time window and the corresponding quality flow rate for that time window.

[0140] This is achieved by simultaneously collecting audio data and known absolute weight data from the feed source (e.g., from a weighing sensor). The weight data is synchronized with the audio data, based on when the feed leaves the storage hopper and the corresponding distance the feed passes through the audio sensor. Note that this distance can theoretically be zero.

[0141] Using synchronized audio and absolute weight data, the audio data is segmented into blocks. Corresponding weight data for each audio segment is interpolated. The difference between the interpolated weight data segments is used to approximate the amount of quality of the section of pipe connected to the audio sensor within that time window.

[0142] This allows for the retrieval of audio data segments corresponding to the mass flow rate within a relevant time window. This data is used as training data for an ML model. The model will learn how vibrations in the pipe correspond to the mass flow rate of the feed. Since the model is trained on learned mass increments for specific time windows, it is learning the mass velocity through the pipe based on pipe vibrations.

[0143] A vibration transducer sensor responds to vibrations passing through its body. Preferably, an audio pickup is used because it has specific acquisition characteristics within certain frequency ranges. It outputs an electrical signal proportional to the vibration amplitude. The output has a waveform shape. An audio pickup, preferably described in the preferred embodiments section above, and... Figure 12 The audio pickup in the housing described herein is positioned to contact the conduit, and the output waveform is derived from various associated vibrations.

[0144] Therefore, in the case of a feeding system, this vibration of the system can be described as follows:

[0145] --Vibration caused by the movement of feed particles or crushed feed clumps scraping against each other.

[0146] --Vibration caused by the movement of the auger in the feed pipe and the scraping of the feed pipe and pellet / crushed feed.

[0147] --Vibration caused by the motor connected to the auger.

[0148] The amplitude and frequency of these vibrations depend on the type, composition, quantity, and velocity of the feed flowing within the pipe. All these combined vibrations are acquired by a transducer and output as waveforms. An audio pickup is then set up to obtain the most critical data for the feed flow application.

[0149] This process consists of an amplifier, an ADC, and custom software. Please refer to [link / reference]. Figure 18 Overview. The amplifier's function is to electrically increase the level of the output waveform of the audio pickup (box 1) (box 2). The ADC converts the amplified waveform's analog electrical values ​​(box 3) into digital values ​​(box 4). The software then converts the digital values ​​from the waveform (box 5) into the original audio file (box 6) suitable for memory storage.

[0150] The extracted waveform is divided into predefined time intervals. These intervals are long enough to ensure that the lowest frequencies can be extracted. The intervals cannot be too long to optimize performance. Figure 17 The process described in step 4 ensures real-time feed flow prediction.

[0151] Storing the raw audio data requires a large amount of memory. Each audio sample is stored in memory along with a timestamp.

[0152] This step involves applying several custom software algorithms designed to extract useful features from the raw audio data. These features must be meaningful for ML propagation, highlighting observed audio variations caused by changes in flow. They also need to have a very small memory footprint. The algorithms are built upon known techniques such as Fast Fourier Transform (FFT), noise cancellation, low-pass filtering, and any other audio feature extraction algorithms related to signal processing.

[0153] Audio functions are stored in memory. The most important audio function in the ML process is acquiring the component frequencies of the waveform. These functions are smaller than the raw audio data (typically 1000 times smaller). This data must be stored in tensor form that can be directly used by the ML software.

[0154] This is the forward propagation algorithm on tensors described in step 3. It outputs a flow rate prediction. The ML model used for forward propagation is built using a large number of audio features appended to the validated flow rate results, such as... Figure 15 As shown. The forward propagation algorithm can be a well-known algorithm in the industry, for example, the system has been successfully tested on the Microsoft Azure Machine Learning Platform.

[0155] Alternatively, online ML forward propagation can be used. In this case, the arrangement described herein provides a wireless connection (as defined in ISO 802.11, 802.15.4, or 802.15.1) or a wired connection for transmitting the extracted audio functionality to the online ML service. Possible options include... Figure 2 As shown.

[0156] This refers to the preloaded (optionally online) ML data model used for forward propagation. It is used for Figure 17 Step 5 of the flow estimation process outlined in the document.

[0157] The traffic prediction calculated in step 5 is timestamped. This value is placed in a local double-buffered memory. Optionally, it can be retrieved by an external component, as described below.

[0158] This value is expressed in kg.s -1 Or Liters.s -1 This indicates feed flow rate. The output value is valid between its appended timestamp and the timestamp of the next output value.

[0159] This embodiment includes a wireless (such as, but not limited to, 802.11, 802.15.4, or 802.15.1) or wired connection for transmitting calculated feed flow rate predictions. Possible options include... Figure 2 As shown.

[0160] Each of the above individual solutions generates accurate flow predictions based on different technologies. The distance-range sensor-based method uses the height of the solid fluid material and, if applicable, the speed of the auger in the pipe, while the acoustic sensor-based method uses pipe vibrations. The calculated measurements are correlated because they are derived from the material flow rate in the same pipe, but they are obtained independently.

[0161] The combined solution results in increased redundancy, reduced uncertainty, and improved accuracy. In particular, combining flow results obtained independently from each method yields a new, significantly less noisy and more accurate calculated feed flow measurement. The sensor combination used in this combined solution allows for the determination of more properties of the flowing solids, such as their density, moisture content, and composition.

[0162] There are many methods available for combining different data sources to improve the accuracy of feed flow forecasting. Standard methods, such as those using Kalman filters, are preferred because they do not require a large amount of computational power and have proven to be very accurate. Furthermore, these methods are very fast, thus enabling them to maintain the same refresh rate as two independent arrangements.

[0163] Alternatively, well-known methods such as voting systems, weighted voting systems, or any other output value integration method using environmental or other relevant data can be used to provide the best solution based on its accuracy and taking into account external conditions and / or feed type.

[0164] The combined approach uses a specific three-stage algorithm to optimize results by placing greater emphasis on reliable individual sensor outputs and eliminating outliers (e.g., from defective sensors).

[0165] This particular embodiment can be generalized and applied to any number of acoustic and lidar sensors. The preferred alternative embodiment described further utilizes two lidar types and one acoustic type of sensor.

[0166] The algorithm consists of three stages, leading to the final refined traffic prediction. The stages are summarized below:

[0167] Scoring: The first stage estimates the reliability of each individual output of each sensor, which is categorized as confidence level. This stage employs as many predicted flows as possible collected from separate processes applied to acoustic and / or distance detection sensor data. These are expressed in kg.s -1 Or Liters.s -1 This indicates that data from its previous state is used in this stage.

[0168] Hybridization: The hybridization stage calculates new confidence levels based on the outputs of the scoring stage and the outputs of previous iterations in the hybridization stage. This stage compares each predicted output from the scoring stage with all other predicted outputs from the scoring stage. These outputs, whether acoustic or lidar outputs, are processed along with all other sensor outputs. The hybridization stage outputs as many new predicted feed flows as permutations of the two inputs. These values ​​are temporary values ​​used as inputs in the next stage.

[0169] Voting: Output for this stage is in Kg.s -1 Or Liters.s -1 This represents the final, accurate feed flow forecast. It uses forecasts and scores from the mixing phase.

[0170] Figure 19 The diagram presents an overview of the data flow. Each stage may include filters or sub-filters that will be described in detail later. Each stage takes the output of the previous stage as its input.

[0171] The scoring phase establishes a confidence score for each sensor prediction (see...). Figure 20(Block 1 in the example). It uses previous predictions to detect outliers. Outliers are defined as values ​​that differ from the signal mean by three standard deviations. In our example, we consider the gradient between the actual value and the previous value to detect outliers. It uses a heuristic about changes in feed flow rate, scoring any value with confidence based on the previous value. Since each sensor has a noisy output, a pre-computed heuristic can be used to avoid over-detecting outliers.

[0172] The output data of the mixing phase consists of the following (the acoustic (A) and distance detection sensors (L1 and L2) have the same structure) (see [link]). Figure 20 Block 2 in the middle:

[0173] The predicted output remains unchanged.

[0174] Additional confidence score (new)

[0175] In the mixing phase, a custom implementation of the Kalman filter is used. They run two processes: calculating new predictions and calculating new confidence scores for them. The new confidence scores are then linked to the predicted values. Figure 21 Each K block in the model is a custom Kalman filter implementation.

[0176] Each new prediction made based on the K block uses the following inputs: one of the predicted feed flow rates as theoretical data, any other predicted flow rate as actual data, and the results of the previous mixing phase as memory data.

[0177] For each new confidence score from K blocks: their inputs are the confidence score from the scoring phase and the score from the previous state. The confidence score is a value that indicates the reliability of the new prediction.

[0178] The output consists of two values, one in Kg.s -1 Or Liters.s -1 The first is the predicted output, and the second is the confidence score (see [link to relevant documentation]). Figure 21 Block 2 in the middle).

[0179] During the voting phase, input data comes from the mixed phase (see...). Figure 22 (Block 1 in the text). The voting machine obtains all predicted values ​​and their confidence scores from the mixing stage and processes variations of the voting algorithm. Confidence scores are used as weights, and values ​​are used as candidates. Custom voting algorithms are the preferred implementation, but other algorithms can also be used. Output values ​​are in Kg.s -1 Or Liters.s -1 This represents the final predicted feed flow rate.

[0180] The arrangements described herein, both independently and in combination, can provide one or more of the following advantages:

[0181] --A method and system for commercial livestock farming for highly repeatable flow measurement.

[0182] --A method and system for providing high-precision near real-time measurement data.

[0183] --A method and system for detecting when a feed system is running without any flowing material in the pipeline. This could be caused by blockages, system leaks, empty silos, feed overlap, or many other common issues.

[0184] --A method and system that can be used in harsh environmental conditions, including those with high levels of dust.

[0185] --A low-cost method and system that can be implemented with low-cost hardware and electronic components.

[0186] --A non-invasive or minimally invasive method and system that does not interfere with feed flow and typically does not contact the feed in the feed tube.

[0187] --A method and system that can be easily installed on feeding systems in existing and new livestock farms.

[0188] --A method and system applicable to any number of locations in the feed pipe to measure the entire livestock shed, herd, or individual animal.

[0189] --A modular approach and system that requires minimal maintenance and can be easily and cost-effectively updated in the field (new sensors, new microcontrollers, new software).

[0190] --A method and system that incorporates wireless communication capabilities and can be easily integrated with other systems used in commercial livestock farming.

[0191] Since various modifications can be made to the invention as described above, and many obviously different embodiments of the invention within the spirit and scope of the claims can also be modified without departing from those spirit and scope, all contents contained in the accompanying specification should be interpreted as illustrative rather than restrictive.

Claims

1. An apparatus for providing an estimate of the flow rate of a flowing solid substance, the apparatus comprising: A roughly horizontal cylindrical pipe; A movable transport element in a pipeline for transporting substances along the pipeline; The movable transmission element includes at least one auger blade surface in the pipe, which rotates about the axis of the pipe; Multiple distance sensors are installed on the upper part of the pipe to detect the distance between the material in the lower part of the pipe and the distance sensors, and generate distance signals to respond to the distance. A control system is provided for analyzing distance signals from the distance sensor over time to provide an estimate of the flow rate.

2. The device according to claim 1, characterized in that, The control system includes an envelope filter that removes sinusoidal artifacts caused by the surface of the at least one auger blade rotating in the pipe.

3. The device according to claim 1, characterized in that, A speed detection sensor is provided for detecting the rotational speed of the surface of the at least one auger blade, and the control system is configured to use the rotational speed to analyze a distance signal from the distance sensor over time to provide a flow rate estimate.

4. The device according to claim 3, characterized in that, The speed detection sensor reacts to the metal on the surface of at least one auger blade present at a location along the pipe.

5. The device according to claim 4, characterized in that, The control system is configured to use the direct derivative of the rotational speed of the surface of the at least one auger blade.

6. The device according to claim 1, characterized in that, The control system is configured to use a volume pipeline mapping algorithm to convert the distance signal from the 2D data of the distance sensor into volume.

7. The device according to any one of claims 1-6, characterized in that, Each of the distance sensors is mounted in a sensor housing carried on the pipeline.

8. The device according to claim 7, characterized in that, A blower is provided to create overpressure in the sensor housing.

9. The device according to claim 8, characterized in that, Overpressure is discharged into the pipe to prevent dust generated by flowing solid matter from flowing into the sensor housing.

10. The device according to any one of claims 1-6, characterized in that, The distance sensor is positioned at different angles around the axis of the pipe.

11. The device according to any one of claims 1-6, characterized in that, The distance sensors are arranged at intervals along the pipe and oriented to detect the distance at those intervals along the pipe.

12. The device according to any one of claims 1-6, characterized in that, The distance sensors are arranged at intervals along the pipe and configured to detect the distance at common locations along the pipe.

13. The device according to any one of claims 1-6, characterized in that, A second estimation system is provided for providing a second estimate of the flow rate, wherein the control system is configured to compare the second estimate with the flow rate estimate to eliminate errors.

14. The device according to claim 13, characterized in that, The second estimation system uses vibration sensors installed on the pipe to respond to vibrations in the pipe.

15. The device according to any one of claims 1-6, characterized in that, It includes a vibration sensor mounted on the pipe, which can respond to vibrations within the pipe, and the vibration sensor provides an input to the control system.

Images