Integrated solid flow meter and method for measuring solid flows
The integrated solid flow meter (MICaS) addresses the challenges of measuring discrete solid flows by using a mechanical and electronic processing system to detect and classify particle impacts, achieving accurate and efficient solid flow rate measurement with reduced errors.
Patent Information
- Application Number
- PCT/PT2024/050017
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-26
- Filing Date
- 2024-05-23
- Publication Date
- 2024-12-05
AI Technical Summary
Current solid flow measurement technologies face challenges in accurately counting and validating discrete solid particle flows due to the high frequency and irregular movement of particles, leading to inefficiencies in data storage and processing.
An integrated solid flow meter (MICaS) comprising a mechanical block with pressurized housings and a membrane, an electronic processing block using a microcontroller and pressure transducers, and a post-processing block with statistical filtering algorithms to detect and classify particle impacts, reducing false positives and negatives.
Enables precise and efficient measurement of solid flow rates with high temporal resolution and reduced error rates, improving the accuracy of solid flow data acquisition and storage.
Smart Images

Figure PT2024050017_05122024_PF_FP_ABST
Abstract
Description
[0001] DESCRIPTION
[0002] " INTEGRATED SOLID FLOW METER AND METHOD FOR MEASURING SOLID FLOWS"
[0003] FIELD OF INVENTION
[0004] The present invention relates to an integrated solid flow meter (MICaS ) and a method for measuring solid flows using said meter .
[0005] BACKGROUND TO THE INVENTION
[0006] The onset of solid particle motion induced by a flow is a classic fluid mechanics problem with many industrial and environmental applications . In studies of the onset of solid transport , in particular, it is important to take precise measurements of the solid flow . At the beginning of the motion, and still far from the generalised motion regime where moving sediments can be regarded as a continuous process , this incipient flow consists of the discrete and irregular movement of particles that must be counted individually for a correct characterisation of the beginning of the movement . It should be noted that the motion of solid particles described in this way can comprise counts of the order of 104particles per second per metre . To achieve statistical robustness for the average particle flow estimator ( solid flow) , the counting process must be carried out over long periods , which results in large amounts of data . The di f ficulty associated with the tasks of identi fying and validating the solid particle count and storing the count data is therefore considerable . This di f ficulty explains the lack of ef fective counting solutions which lead to solid flow measurement solutions that are at once economical in terms of data storage volume and ef fective in terms of the counting process . This invention is motivated by and addresses the di f ficulties in measuring discrete solid flows .
[0007] US5986553A describes a flow meter that measures the flow of solids , in particular a Doppler flow meter, and which comprises a flow tube , a sensor and an indicator . The sensor is placed in a sensor tube and as the product passes through the sensor signal ( low microwave energy) , that energy when it comes into contact with the particles undergoes a Doppler shi ft which is detected by the sensor . The width of the sensor beam must be at least the diameter of the flow tube , so that it covers the entire cross-sectional area of the flow tube . All particulate materials flowing through the flow tube come into contact with the beam and are thus reflected . The instrument is calibrated and the algorithms determined from the calibration to make the indicator provide the best fit function in the indicator ' s output zone . The algorithm converts the solids flow signal into total solids flow .
[0008] US2008148874A1 relates to a solid particle counting system for measuring solid particle number concentrations in the exhaust pipe of the engine or vehicle in real time and includes an evaporator unit , a diluter arrangement and a particle counter . The diluter arrangement mixes the dilution gas with the gases in the evaporation unit at a dilution ratio , and the diluter arrangement includes a gas flow meter located upstream of the evaporation unit to measure the flow of gases in the evaporation unit, which in turn flows into the diluter arrangement.
[0009] CN103541798A relates to a detection method based on a solids flow meter for the amount of ammonia remaining in a solid ammonia storage system. The solid ammonia flow meter is arranged in an ammonia storage tank and is connected in series to an ammonia distribution pipe and a display controller. In turn, a pressure sensor mounted on the ammonia storage tank is connected to a temperature sensor, an ammonia mass flow meter and a DCU (Domain Control Unit) , and the DCU and display controller are connected to an ECU (Electronic Control Unit) .
[0010] BRIEF DESCRIPTION OF THE DESIGNS
[0011] The following is a detailed description of the invention with reference to the attached drawings, in which:
[0012] Figure 1 shows the MICaS diagram, indicating the separator plates (1) between each pressurised housing (2) and the stop bar (3) . The particles (4) fall onto the membrane (5) of a pressurised housing, inducing a pressure variation inside. To ensure that the particles with the greatest amount of movement do not fly over the pressurised housings, a stop bar (3) is placed in the channel to force all the particles to reach the pressurised housing.
[0013] Figure 2 shows a schematic of all the blocks in the
[0014] MICaS system:
[0015] (A) mechanical block: separator plate (1) , pressurised housing 1 (2) , stop bar (3) , particle (4) , membrane ( 5 ) ; (B) pneumatic circuit: pump (6) , VG (7) , Vi (8) , pressure transducer (Ti) (9) ;
[0016] (C) the electronic block: Pressure transducer (Ti) (9) , Microcontroller (10) ;
[0017] (D) Post-processing block: Post-processing (11) .
[0018] Figure 3 illustrates the layout of the pressurised housings in different views.
[0019] Figure 4 shows a schematic of the lateral distribution of the pressurised housings in the channel: (a) housings with uniform dimensions; (b) housings with non-uniform dimensions .
[0020] Figure 5 illustrates a schematic of the separator plate in different views.
[0021] Figure 6 shows a schematic of the pressurised housing flanked by the separator plates in different views.
[0022] Figure 7 shows diagrams of the pressurised housings and separator plates installed in a channel.
[0023] Figure 8 illustrates a schematic of the pneumatic circuit. Each pressurised housing Ci (i=l,2,3) connects to its transducer, Ti (i=l,2,3) , and to the pressure element, P, which will impose the desired pressure. A set of valves, Vi (i=l, 2, 3) , isolates each pressurised housing. The VG valve is the valve for the general pressurisation circuit.
[0024] Figure 9 shows the MICaS electronic block. Figure 10 shows the flowchart of the operations performed by the electronic block and the firmware. The signal x[n] from each event n (Step [1] in Figure 10) is filtered by the infinite impulse response (HR) Butterworth filter (Step [2] ) - From the result yO [n] , the energy associated with each event is determined (Step [3] ) . This energy is used by the k-NN (k-Nearest Neighbour) algorithm (Step [4] ) to classify the energy window, the energy level and the class to which the event belongs, namely: Turbulence, Silence and Impact. This information is used by the contiguous class Filter (Step [5] ) which determines whether it is a new impact if the following conditions are met: temporal separation of the detection class is greater than or equal to the minimum separation count, and if the contiguous counts of the impact class are greater than or equal to the minimum impact count. The result of step [5] is the classification of the event into Impact or Non-Impact, the energy associated with the event, and the time at which it occurred. This information is fed to a statistical impact filter (Step [6] ) which validates the impact if: the time interval between the impact instant t (n) and the previous impact instant t(n-l) and the ratio between the energy of impact n, E (n) , and previous impact n-1, E(n-l) , do not correspond to the confidence intervals of the aftershocks or bounces. Otherwise, the event is rejected. Information on validated impacts is displayed on the LCD (Liquid Crystal Display) or sent via USB (Universal Serial Bus) or serial (RS-232) ports (Step [7] ) . The information after the Butterworth filter is also sent to the LCD display and to the USB and RS-232 serial ports (Recommended Standard 232) .
[0025] Figure 11 shows a map of false positives obtained from an experimental MICaS test. This figure shows a map of local empirical probabilities of false positives , determined on the basis of the impacts recorded in the pressurised housings when cross-referenced with the respective video recording ( example of an experimental test ) . To obtain the 2D histogram, energy-interval ratio for false positives , a video recording is first made of the initial n minutes of MICaS operation (where n is adj usted according to the solid flow rate , to allow for some statistical representativeness ) , followed by a manual classi fication of false positives in a spreadsheet by cross-referencing the complete event records produced by MICaS (which must include the energy associated with each detected event ) and the video events for each membrane housing . Finally, a computer code is run that imports the data from the spreadsheet and then produces the 2D histogram mentioned above , which relates the ratio of energies and time intervals between events to false positives . Therefore , whenever MICaS records an impact event and the video indicates a situation of aftershock or bouncing of the sphere , situations that correspond to false positives , these are classi fied in the spreadsheet , as are the real impacts , identi fying the membrane housing where the event occurred, the video time instant , the respective MICaS event number with the associated energy ratio (between the event that occurred and the previous event detected in the same membrane housing, whether this is a false positive or not ) . It should be noted that the aftershock situation corresponds to the propagation of a pressure wave that causes the membrane of a housing to vibrate, resulting in a false impact detection, as no impact actually occurred . The bounce situation, on the other hand, consists of repeated successive impacts on the housing-membrane of the same sphere , for which
[0026] MICaS generated an impact detection event . The 2D histogram produced relates the time interval s between events with the energy ratios also between successive events and in the same membrane housing, and makes it possible to empirically observe the trend of the statistical distribution of false positive events , in relation to the two parameters mentioned . Each peak in the 2D histogram indicates the local probability with which a false positive occurs for a given pair of energy ratio values and time interval between successive events .
[0027] SUMMARY OF THE INVENTION
[0028] The present invention relates to an integrated solid flow meter comprising a mechanical block, an electronic processing block and a post-processing block in which :
[0029] - the mechanical block comprises one or more pressurised housings flanked by one or more separator plates , a membrane covering an upper part of the pressurised housings , a pneumatic circuit associated with the said pressurised housings and a stop bar ; the electronic processing block comprises a microcontroller, which runs firmware , and a pres sure transducer ; and
[0030] - the post-processing block comprises the application of a statistical filtering algorithm implemented in a software program running on a CPU ( central processing unit ) .
[0031] In one form of embodiment of the present invention, the separator plates of the integrated solid flow meter comprise a spur on the upper front part . In another form, the pneumatic circuit comprises each pressurised housing connected to its transducer and pressure element , a set of valves that isolate each pressurised housing, and a general pressurisation valve .
[0032] In one aspect of the present invention, the microprocessor comprises an output for a portable LCD ( liquid crystal display) screen, outputs for USB and serial ports .
[0033] In one embodiment of the present invention, the integrated solid flow meter also comprises multiplexers for processing signals from the di f ferent pressurised housings .
[0034] In one aspect of the present invention, the membrane is made of a rubber-like material .
[0035] In another aspect of the invention, the pressurised housings are made of polyethylene terephthalate glycol ( PETG) .
[0036] In one design, the pressurised housings are the same width .
[0037] In another embodiment of the present invention, the pressurised housings vary in si ze .
[0038] The present invention also relates to a method for measuring solid flows , using said integrated solid flow meter, which comprises the following steps :
[0039] - detection of the impact of solid particles from the solid flow on the membrane by the microcontroller ; - measuring variations in pressure distribution caused by the impact of these solid particles on the pressure transducer ;
[0040] - conversion of the pressure signal into an analogue electrical signal by means of the pressure transducer ;
[0041] - conversion of the analogue electrical signal into a digital electrical signal in the microprocessor ;
[0042] - filtering the digital electrical signal through a low-pass filter, with subsequent ampli fication and filtering again through an anti-aliasing filter ; application of a statistical filter using an algorithm;
[0043] - presentation of results on a screen or sound device and exporting data to a computer .
[0044] In one form of embodiment of the present invention, the stage of detecting the impact of solid particles on the membrane using the microcontroller comprises the following steps :
[0045] - filtering the digital signal using an HR Butterworth filter ;
[0046] - discriminating impact based on analysing the energy associated with the detected frequencies resulting from the particle ' s impact with the membrane ;
[0047] - identi fying impacts using the k-NN algorithm, where k=3 ;
[0048] - analysing a set of candidate particle events ;
[0049] - applying a statistical filter to discriminate between three types of impact events defined as impacts , bounces and aftershocks .
[0050] DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an integrated solid flow meter (MICaS ) .
[0051] The integrated solid flow meter of the present invention consists of a device for measuring solid flows comprising three main blocks : a mechanical block, an electronic processing block and a post-processing block .
[0052] The meter works based on the conversion of particle impacts in a pressurised housing ( impact device ) , which result in a pressure variation, which is converted into an electrical signal by means of pressure transducers . The electrical signal is then processed and analysed by firmware- controlled electronics and is post-processed using software programs designed for the purpose .
[0053] This device allows uninterrupted measurement of solid flow rates with high temporal resolution .
[0054] The three main blocks of the meter of the present invention are described below .
[0055] 1 . Mechanical block
[0056] The mechanical block is located in a channel where particles are flowing, in which the flow of solids is to be measured . The mechanical block comprises one or more pressurised housings , an associated pneumatic circuit and a membrane . A pressurised housing can be seen as a drum, where the spherical particles hitting the membrane that covers the top of the housings generate changes in the pressure distribution inside the pressurised housing, and these will be transmitted to pressure transducers. The resonance frequency will depend on the material of the membrane, its mass and its dimensions (Garret, S.L. (2017) . Understanding acoustics - An experimentalist's view of acoustics and vibrations. Springer. DOI: 10.1007 / 978-3-319-49978-9; Leissa, A.W. (1973) . The free vibration of rectangular plates. Journal of Sound Vibration, 31, pp . 257-293; Blevins, Robert D. (1979) . Formulas for natural frequency and mode shape, Van Nostrand Reinhold Company. New York. ISBN: 0-442- 20710-7) .
[0057] A generic pressurised housing is illustrated in Figure 3.
[0058] The pressurised housings are made from polyethylene terephthalate glycol (PETG) and the membrane covering the top of the housings is made from a rubber-like material.
[0059] The housings can be placed in the channel in different ways and can be adapted to the width of the channel, thus making it possible to discretise the flow of particles in the channel. As an example, some possible configurations are illustrated in Figure 4 (a) , with pressurised housings of uniform width and Figure 4 (b) with pressurised housings of varying dimensions. The first case allows for uniform discretisation of the particle flow, while the latter allows for a non-uniform description of the particle flow, allowing certain parts of the channel to have greater or lesser spatial resolution depending on the need, for example if the distribution of the solid flow is uneven. The sensitivity of the membrane to the impact can be regulated by the choice of the membrane material and adjusting the pressure in the housings.
[0060] The housings should be placed at a lower level than the bed where the solid flow occurs, so that the particles fall onto the membrane. This configuration can lead to the appearance of a recirculation zone and consequent turbulent structures of significant dimensions in the liquid medium. To reduce the effect of the turbulent structures on the membrane surfaces of the pressurised housings, effectively reducing the size of these structures and therefore the magnitude of the turbulent noise, the housings should be separated by plates placed in the longitudinal direction of the channel, as shown in Figures 5 and 6.
[0061] The objectives of these separator plates are: (a) to reduce the influence in the liquid medium of possible large turbulent structures which, by generating noise, could cause false detections of particle impacts; (b) to prevent particles impacting one housing from bouncing off and impacting the neighbouring housing, and (c) to reduce the propagation of vibrations, by solid medium, between adjacent housings .
[0062] The separator plates are also fitted with a spur on the upper front part, with the threefold aim of: (i) guiding the particles to the central, more sensitive area of the membrane, (ii) preventing the formation of separation zones that would occur if the spur did not have a hydrodynamic shape in all three dimensions, and (iii) preventing interference in the solid flow due to the intermittent blocking of the particles captured in the hydrodynamic separation zone .
[0063] The separator plates are also appropriately si zed to accommodate the stop bar . This stop bar is placed downstream of the separator plates , immediately after the drop where the pressurised housings are accommodated ( Figure 3 ) . The function of this plate is to stop the particles that are travelling with the greatest amount of movement and make them fall onto the membranes of the pressurised housings .
[0064] The introduction of the spur separator plates has led to a signi ficant improvement in the device ' s performance , reducing false positives and false negatives with respect to the version initially tested .
[0065] Figure 6 shows a housing encircled by the separator plates and Figure 7 shows the installation of the housings along the width of the channel .
[0066] A pneumatic circuit was built with valves allowing the isolation of each of the pressure housings . One of the piezometric ori fices connects directly to the transducer while the other connects to the supply circuit , as shown in Figure 8 .
[0067] 2 . Electronic processing block
[0068] The electronic processing block is built based on the STM32 F4 microcontroller . This microcontroller has an output for a portable LCD screen so that results can be seen in real time , as well as outputs for USB and Serial ports , allowing connection to a computer. The electronic processing block is shown in Figure 9.
[0069] The pressure signal is converted into an electrical signal by means of a transducer, such as the NXP MPXV5004G transducer. The pressure signal from the transducers corresponds to the absolute pressure. The pressure difference between consecutive pressure housings is calculated using a differential amplifier. This makes it possible to compensate for any effects of turbulence on the pressure signal; however, impact detection is only carried out based on absolute pressure signals.
[0070] This analogue electrical signal is first filtered by a low-pass filter to remove high-frequency noise. The resulting signal is then amplified and filtered again by an anti-aliasing filter of order n (e.g. n = 8) . These filters are part of the anti-aliasing filtering system in the step before the conversion of the analogue signal into a digital signal, but by themselves they cannot completely prevent anti-aliasing. The analogue-to-digital conversion is performed on the ADC (Analogue to Digital Converter) input of the microcontroller for further processing.
[0071] Almost simultaneous processing of the signals coming from the different pressurised housings can be achieved through the use of multiplexers. The absolute and differential pressure signals are connected to the inputs of the MUX1, MUX2 and MUX3 multiplexers, as well as additional test signals, totalling 3 x 8 = 24 signals. The multiplexers make it possible to reduce the cost of the hardware, thereby eliminating the need to include programmable gain amplifier blocks, low-pass filters, and ADC units per channel. On the downside, when multiplexers are not used, there is an increase in inter-channel interference, but this does not affect the operation of the device in any problematic way.
[0072] 2.1 The firmware
[0073] The firmware that runs on the microcontroller controls, among other things, the multiplexers, the selection of the pressure signal to be analysed in a circular pattern (round robin) .
[0074] The microcontroller controls the programmable gain unit (PGA) , with a gain that varies between 0.35 (attenuation) and 17 (amplification) . The PGA' s function is to condition the pressure signal so that its dynamic range is adjusted to the dynamic range of the ADC (analogue to digital converter) , for greater efficiency in terms of resolution. This avoids saturation or clipping of the original small or large amplitude signal. The PGA unit treats the DC (direct current) component of the signal separately from the AC (alternating current) component, allowing different gains for each of these components.
[0075] Impact detection is carried out in the microcontroller to enable real-time analysis. Impact detection is carried out in the following stages:
[0076] 1. filtering of the digital signal using a Butterworth filter implemented in the form of an HR Butterworth filter to remove the frequencies dominated by turbulence and noise;
[0077] 2. discrimination of impacts based on analysing the energy associated with the detected frequencies resulting from the particle's impact with the membrane because :
[0078] 2.1 an impact of the particle on the membrane results in a pressure signal with a periodic component containing several frequencies;
[0079] 2.2 each frequency has its own associated energy to be determined by Fourier analysis;
[0080] 2.3 the pair (frequency, energy) is used to classify events;
[0081] 3. impacts are identified using the k-NN algorithm, where k=3, i.e. the possible events are: (a) silence, (b) particle candidate and (c) turbulence;
[0082] 4. analysing all the events that are candidates for particles, in order to distinguish between the main impact and the bounce;
[0083] 5. Discrimination of impact events into three types, impacts, bounces and aftershocks, by applying a statistical filter (Step 5 in the diagram in Figure 10.
[0084] In Step 1, the experimental analysis concluded that the frequency range of interest was between 150 Hz and 400 Hz.
[0085] A detailed analysis showed that not all impacts corresponded to particles. For example, the same particle can impact the membrane several times: the main impact and the bounce, producing two signals in which the second, called the bounce, is a false positive. It was also observed that an impact in housing 1 could cause a false positive in adjacent housings (i+1 and 1-1) , which prompted the introduction of the separator plate described above.
[0086] The filter in Step 5 consists of a choice operation in the plane defined by two variables: the time interval (T*) and the energy ratio (E*) between the candidate event and the last validated event. In the case of aftershocks, it should be noted that the last impact may have been recorded in adjacent pressurised housings. It has also been observed that bounces and aftershocks occur in a very closed region of the (T*, E*) plane. Therefore, candidate impacts whose (T*,E*) values were in the aforementioned closed region were eliminated. The boundaries of the region were determined experimentally. These boundaries are determined on the basis of the map of local empirical probabilities of false positives in Figure 10. This map is obtained by calculating the joint empirical probability of false positives in each element AT*,AE* of the space (T*,E*) .
[0087] The k-NN algorithm is a supervised learning algorithm, so it needs to be trained in order to achieve a useful level of performance, and it cannot work autonomously without going through an initial learning phase. To this end, there are two stages that form part of supervised learning, i.e. a training stage followed by a validation stage. Two sets of data are also required, as disjoined as possible, one for each of these stages, respectively. Each element of the training and validation sets consists of a recording of the pressure sensor values over a given period, a window of samples in which a given event occurs and the corresponding classification of the event provided by a human operator. In this way, the k-NN algorithm is fed the training data and the respective classifications of each recorded event, resulting in the learning of a weight table for each event class. The performance of the k-NN algorithm with the learnt weight table is then validated with the disjoined validation data, measuring the degree of effectiveness of detection in percentage of correct detections. If the detection ef ficiency is lower than desired, a more appropriate window si ze or frequency range should be sought for the pressure signals obtained from the membrane housings , and a new training and validation stage should be carried out . Once the desired ef ficiency has been achieved, the training process is complete and MICaS can switch to autonomous operation mode .
[0088] The sets of events for training the k-NN algorithm can be obtained in di f ferent ways : ( a) out of the water, by dropping particles in free fall onto the membranes , and recording the di f ferent types of impact ; (b ) under hydrostatic conditions , by dropping the particles onto the membrane ; ( c ) under conditions similar to those of the flow, by releasing some test particles into the channel and letting them be transported by the flow until they reach the membrane , while s imultaneously recording the impacts with a video camera . This procedure can be quite time-consuming, as the k-NN algorithm needs to be trained for di f ferent types of impact and corresponding signals .
[0089] In this case , the data acquisition process for training consists of two main stages , one without particles to determine the ef fect of flow turbulence and background noise , and the second taking measurements under cross-operating conditions with video recording and recording the impacts and pressure variations in the housings .
[0090] The recorded and categorised pressure signals are then divided into isolated blocks of short duration with the pressure variations corresponding to the respective category : impact , silence or turbulence / noise . The files with the categorised pressure signals are then divided into two sets , the training set and the validation set , which are disj oined sets . The training data set is analysed by a software program that will extract the parameters of the k-NN algorithm, which will be configured in the MICaS firmware . Using the same software program, the performance of the training data learnt by the k-NN algorithm is analysed from a statistical point of view about the validation data, and the ef fectiveness of the detection of the various categories is observed . Once the desired level of ef ficiency for detecting the various categories has been reached, the k-NN parameters resulting from the training are then configured in MICaS . The k-NN works as a function of the accumulated energies of the pressure signal over a series of time intervals and for one or more frequency ranges , the lower case being a single time interval . The MICaS training software program and firmware also support a related method of k-NN, namely LMNN ( large margin nearest neighbour ) , which allows the spatial operating region of k-NN to be deformed in order to improve detection ef ficiency by reducing the spatial collision between elements of di f ferent categories . In the experimental tests , only the base k-NN was used, but support for LMNN is also implemented . There is also a postprocessing tool that allows a statistical filter matrix to be derived semi-automatically, based on the local empirical probability of false positives , as described above , from a file with MICaS events categorised by a manual operator and the corresponding events generated automatically by MICaS accompanied by their energy values . In this way it is possible to carry out a real-time measurement directly with MICaS , with lower precision, and then calibrate the acquired data and obtain a more accurate solid flow measurement afterwards . 3 . Post-processing block
[0091] As mentioned above , the results obtained in real time with MICaS allow the lateral distribution and time series of the solid flow to be analysed quickly and easily, but there is an error rate which usually overestimates the solid flow . This error can be substantially reduced by post-processing with the application of the statistical filter that uses the calibration parameters previously obtained .
[0092] We now describe a set of post-processing algorithms that assume the possibility of long-term acquisitions with MICaS , but sharing the initial acquisition of impacts simultaneously with the video acquisition . The video should make it possible to identi fy the approximate instants when the particles hit the membrane of the housings , while also allowing the hous ing to be identi f ied . In the tests carried out , an initial j oint recording overlap of 5 minutes was used with a video camera at 300 images per second and simultaneous MICaS acquisition in all the membrane housings . For low solid flow rates , it may be necessary to increase the duration of the j oint acquisition in order to acquire statistically representative calibration data . Under these conditions , information is then collected by visual inspection with the approximate times of the impacts and the respective housing / membrane where they occurred . The relationship between possible bounces is also considered, retaining an acyclic graph with a link to the previous or initial impact .
[0093] The post-processing algorithm or generalised statistical filtering algorithm ( GSA) comprises : loading the MICaS detection time series corresponding to the acquisition period simultaneously with the video recording;
[0094] • loading information with instants of impacts on the video confirmed by visual / manual analysis of the calibration period;
[0095] • applying an algorithm for cross-referencing MICaS detection data with video calibration data ( see Algorithm A) ; this algorithm is designed to load additional information from MICaS with information on the impact energy and instant of detection;
[0096] • for each housing with MICaS events :
[0097] • get all the MICaS events from that housing;
[0098] • get all the video events from that housing;
[0099] • for each MICaS impact event : calculate the energy ratio between the current impact event and the previous one from the same housing;
[0100] • calculate the time elapsed between the previous impact of the same housing and the current impact event ;
[0101] • i f the MICaS event has a corresponding video event ( see Algorithm B ) :
[0102] • add the n-tuple to the results with correspondence : obj ect instance with MICaS event , time interval and energy ratio for the previous event ; • otherwise, if the event still corresponds to the MICaS and Video simultaneous recording period:
[0103] • add the entry to the unmatched results: the MICaS event n-tuple, time interval and energy ratio for the previous event;
[0104] • load all MICaS events from the entire MICaS acquisition period (long acquisition) with or without Video matching;
[0105] • for each event loaded:
[0106] • if there is no previous event:
[0107] • add the event to the results;
[0108] • otherwise, (there are previous events) :
[0109] • calculate the energy ratio;
[0110] • calculate the time interval for the previous event ;
[0111] • using the global statistical filter, determine whether or not this event should be considered a false positive (see Algorithm D) - based on the energy ratio and time interval obtained;
[0112] • if the statistical filter indicates that the event is not a false positive:
[0113] • add the event to the results;
[0114] • otherwise, (false positive) :
[0115] • discard / ignore the event; sort the results by time of occurrence; export the filtered MICaS detections.
[0116] Algorithm A refers to the automatic cross-referencing of MICaS / video data and comprises:
[0117] • for each MICaS detection in a given membrane housing:
[0118] • as long as there are video events for this membrane housing:
[0119] • search for the matching event from the manual video analysis - video and MICaS events less than 50 ms apart:
[0120] • when there is a match between the two events :
[0121] • associate the instant of occurrence of the MICaS event with the video event;
[0122] • mark the MICaS event (object) as matching;
[0123] • add the video event to the list of matching events;
[0124] • move on to the next MICaS event;
[0125] • go to the next video event;
[0126] • otherwise, (no match) :
[0127] • if the video event precedes the MICaS event by more than 50 ms:
[0128] • go to the next video event; otherwise :
[0129] • move on to the next MICaS event ;
[0130] • for each housing with MICaS events :
[0131] • for each video event in the respective housing :
[0132] • i f the video event does not have an associated MICaS event and i f it is not a bounce :
[0133] • put the event on the list of missed impacts ( false negatives ) ;
[0134] • return the impact events with a match, the missed impact events and the events without a match .
[0135] When searching for the matching event from the manual video analysis , where the video and MICaS events are less than 50 ms apart , it is normal for there to be a small variability between the two instants (video and MICaS ) , due to the lower precision of the time count based on the image frame number in the video record and also because of the propagation time of the pressure waves after impact and the processing time of the detection code in the MICaS device firmware .
[0136] Algorithm B refers to determining the global histograms of matching and non-matching events and determining the false positive and false negative tables :
[0137] • for each housing with events :
[0138] • i f it ' s the first or last housing, skip it and move on to the next housing; • add to the global events ( all housings ) the events with and without video matches from the present housing, but without including histogram information;
[0139] • calculate the 2D histogram of the global events with matches , placing the time intervals on the x-axis and the energy ratios on the y-axis - calculate with 50 intervals for each axis ;
[0140] • calculate the 2D histogram of global events without matches , placing the time intervals on the x-axis and the energy ratios on the y-axis , using the same intervals resulting from the application of the previous 2D histogram;
[0141] • calculate the global false positives and false negatives (Algorithm C ) ;
[0142] • export the statistical filtering maps .
[0143] Algorithm C refers to the determination of the table / map representing the local empirical probability o f false negatives and false positives :
[0144] • create a two-dimensional map for false positives with the same dimensions as the 2D histograms with and without a match .
[0145] • for each entry in the two-dimensional array :
[0146] • i f there are matching events , or i f there are nonmatching events , or both :
[0147] • the false positive entry corresponding to the same two-dimensional position in the 2D histograms is calculated : number of MICaS hits without a match / (number of MICaS hits with a match + number of MICaS hits without a match) ;
[0148] • the false negative entry corresponding to the same two-dimensional position in the 2D histograms is calculated : number of hits Video without match / (number of hits Video with match + number of hits Video without match) ;
[0149] • return tables / maps with the local empirical probability of false positives and false negatives .
[0150] Algorithm D refers to carrying out a Bernoulli test with the appropriate probability :
[0151] • find the x and y index of the map entry with the local empirical probability of false positives ( obtained from Algorithm C ) ; to do this , use the histogram interval division for the time interval (x ) and the energy ratio ( y) ;
[0152] • i f the index obtained for any of the dimensions is outside the range , use the first or last valid index, whichever is closest to the desired value ;
[0153] • apply a Bernoulli test with uni form distribution and the probability value obtained from the false positive table ;
[0154] • generate a random number between 0 and 1 with a uni form distribution;
[0155] • i f the random value is above the false positive probability value : consider the event valid;
[0156] • otherwise (value equal to or below) :
[0157] • consider the event to be a false positive;
[0158] • return the result of the Bernoulli test (true or false) .
[0159] In the context of this description, the term "comprising" should be understood as "including but not limited to". As such, it should not be interpreted as "consisting only of".
[0160] It should be noted that any X value presented in the course of this description should be interpreted as an approximate value of the real X value, since such an approximation to the real value would reasonably be expected by the specialist in the technique due to experimental and / or measurement conditions that introduce deviations from the real value. So, for example, the reference to "80%" means a value of "around 80%".
[0161] Unless otherwise stated, the value ranges given in this description are intended to provide a simplified and technically accepted way of indicating each individual value within its range. By way of example, the expression "15% to 25%" or "between 15% and 25%" means any value within this range, for example 15%; 16%; 17%; 18%; 19%; 20%; 21%; 22%; 23%; 24%; 25%; or others, for example including a decimal place . It should be noted that although the present invention has been described with reference to its preferred forms of embodiment , many modi fications and alternatives can be made by a person skilled in the art without leaving the scope of the invention, which is defined by the claims .
Claims
CLAIMS1 . An integrated solid flow meter characterised by comprising a mechanical block, an electronic processing block and a post-processing block in which :- the mechanical block comprises one or more pressurised housings , flanked by one or more separator plates , a membrane covering an upper part of the pressurised housings , a pneumatic circuit associated with the said pressurised housings and a stop bar ; the electronic processing block comprises a microcontroller, which runs firmware , and a pres sure transducer ; and- the post-processing block comprises the application of a statistical filtering algorithm implemented in a software program running on a CPU ( central processing unit ) .2 . The integrated solid flow meter according to Claim 1 , characterised in that the separator plates comprise a spur on the upper front part .3 . The integrated solid flow meter according to any of the preceding claims , characterised in that the pneumatic circuit comprises each pressurised housing connected to the respective transducer and pressure element, a set o f valvesisolating each pressurised housing, and also a general pressurisation valve .4 . The integrated solid flow meter according to any of the preceding claims , characterised in that the microprocessor comprises an output for a portable LCD screen, outputs for USB (universal serial bus ) and serial ports .5 . The integrated solid flow meter according to any of the preceding claims , characterised in that it further comprises multiplexers for processing signals from the di f ferent pressurised housings .6 . The integrated solid flow meter according to any of the preceding claims , characterised in that the membrane is made of a rubber-type material .7 . The integrated solid flow meter according to any of the preceding claims , characterised in that the pressurised housings are made of polyethylene terephthalate glycol ( PETG) .8 . The integrated solid flow meter according to any of the preceding claims , characterised in that the pressurised housings have the same width .9 . The integrated solid flow meter according to any one of Claims 1 to 7 , characterised in that the pressurised housings have variable dimensions .10 . A method for measuring solid f lows using the integrated solid flow meter of any of the preceding claims , characterised in that it comprises the following steps :- detection by the microcontroller of the impact of solid particles from the solid flow on the membrane ;- measurement of variations in pressure distribution caused by the impact of these solid particles on the pressure transducer ;- conversion of the pressure signal into an analogue electrical signal by means of the pressure transducer ;- conversion of the analogue electrical signal into a digital electrical signal in the microprocessor ;- filtering of the digital electrical signal through a low-pass filter, with subsequent ampli fication and filtering again through an anti-aliasing filter ; application of a statistical filter using an algorithm;- presentation of results on a screen or sound device and export of data to a computer .11 . The method for measuring solid flows according to Claim 10 , characterised in that the step of detecting the impact of the solid particles on the membrane via the microcontroller comprises the following steps :- filtering the digital signal using an infinite impulse response Butterworth filter ;- discriminating impact based on analysing the energy associated with the detected frequencies resulting from the particle ' s impact on the membrane ; identi fying impacts using the k-NN ( k-Nearest Neighbour ) algorithm, where k=3 ;- analysing a set of candidate particle events ;- applying a statistical filter to discriminate between three types of impact events defined as impacts , bounces and aftershocks .
Citation Information
Patent Citations
Mass flow meter based detection method for remaining ammonia amount of solid ammonia storage system
CN103541798A
Solid particle counting system with flow meter upstream of evaporation unit
US20080148874A1
Flow meter that measures solid particulate flow
US5986553A
Apparatus for sensing moving particles or small moving objects
GB1384882A
Particulate matter impact sensor
US20160025531A1
