In-process adjustment method for crushing system

Abstract

The goal is to predict the properties of particles. A method, system, and apparatus, including a computer program encoded on a computer storage medium, for in-process adjustments to a crushing system, the method including: acquiring pre-crush particle data indicative of characteristics of a portion of particles prior to crushing the portion of particles with a crushing system, acquiring configuration data indicative of one or more settings of the crushing system, acquiring post-crush particle data indicative of characteristics of the portion of particles after crushing the portion of particles, training a predictive model using the pre-crush particle data, the configuration data, and the post-crush particle data, the training including processing the pre-crush particle data and the configuration data using the predictive model to obtain a corresponding output including predicted characteristics of the portion of particles after crushing the portion of particles with the crushing system, and adjusting parameters of the predictive model based on comparing the output of the predictive model to the post-crush particle data.

Description

[Technical Field]

[0001] The present specification relates generally to rock aggregate crushing processes. [Background technology]

[0002] Billions of tons of aggregates are used worldwide each year in the construction of buildings, roads, and other critical infrastructure. Much of the aggregate consists of crushed stone produced in surface quarries. Quarries vary significantly in age and production, with some producing over 10 million tons of aggregate annually. In quarrying operations, aggregate is produced from ore bodies by excavating, dredging, or blasting. The output aggregate is often then loaded onto a conveyor belt, where it is transported to crushing equipment. Aggregates output from quarries can suffer from inconsistencies between batches. In some cases, the distribution of output product differs significantly from market demand specifications, leading to waste and costs associated with storing and retrieving low-demand product.

[0003] Collecting real-time data on the output of a rock crusher can be difficult due to the scale and speed of operation. With thousands of tons of aggregate moving through a quarry system daily, it is desirable to perform comprehensive testing of the output aggregate while maintaining high productivity. Collecting size and shape data for individual particles of aggregate requires large amounts of data to be collected and stored. Real-time or in-process data analysis is required to implement operational changes to alter the distribution of the output aggregate to align with the target output. Summary of the Invention

[0004] Generally, the present disclosure relates to processes and systems for in-process adjustments to a crushing system. A predictive model can be trained to predict post-crushing properties of aggregate particles undergoing a crushing process. For example, the predictive model can receive as input data indicative of geometric and chemical properties of particles crushed by a rock crushing system. The input can include, for example, particle characterizations generated by a particle characterization model. The predictive model can also receive as input data indicative of settings of the rock crushing system. The predictive model can process the input data to generate output data including predicted properties of the particles after crushing. Parameters of the predictive model can be adjusted based on comparing the predicted properties to measured properties of the particles after crushing. In some examples, the measured properties of the particles are determined by providing low-resolution or low-fidelity sensor data to the characterization model. After training, the predictive model can be used to predict the post-crushing properties of the particles.

[0005] The estimated characterizations of the particles output by the characterization model can be used to adjust parameters of the process for crushing the aggregate. For example, the computing system can determine an error by comparing the estimated properties of the batch of crushed aggregate particles output by the crushing system with target properties. In some examples, the estimated properties can be determined using low-resolution sensor data processed by the characterization model. Based on the error, the computing system can perform a feedback control process to adjust the settings of the crushing system.

[0006] In some examples, the computing system can determine the error by comparing predicted post-crush properties of the batch of crushed aggregate particles to target properties. In some examples, the predicted post-crush properties may be output by a predictive model. Based on the error, the computing system can execute a feedforward control process to adjust the settings of the crushing system.

[0007] Sensor data from the input aggregate, the output aggregate, or both can be synchronized with the crushing system's operating settings, such as data conveyor belt speed, cone rotation speed, cone distance, material feed rate, work surface opening size, crusher operating speed, and other settings.

[0008] The sensor data and operational settings may be provided as input to a model or algorithm that may be used to continuously update the operational settings to achieve a desired distribution of output aggregate. In some implementations, the model may be integrated with the crushing system equipment to continuously optimize the operational settings in an automated manner.

[0009] The subject matter described herein can be implemented in various implementations and may provide one or more of the following advantages: The disclosed system can provide high-throughput, accurate characterization of large volumes of aggregate particles. The system can provide real-time adjustment of crushing system parameters during the crushing process. The disclosed techniques can be implemented to increase and improve the uniformity of crushed particles output by the crushing system. The disclosed techniques can be implemented to align crushed particle properties with target properties. The target properties can be updated over time, for example, based on user input. The disclosed techniques can improve crushing system efficiency and reduce waste by improving the quality of the output crushed particles. The disclosed techniques can be applied to the characterization and process optimization of any mined or recycled aggregate that undergoes a crushing process.

[0010] The present disclosure also provides a computer-readable storage medium coupled to one or more processors and storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations in accordance with implementations of the methods provided herein.

[0011] The present disclosure further provides a system for implementing the methods provided herein, the system including one or more processors and a computer-readable storage medium coupled to the one or more processors and storing instructions, the instructions, when executed by the one or more processors, causing the one or more processors to perform operations according to an implementation of the methods provided herein.

[0012] Other implementations of the above aspects include corresponding systems, apparatus, and computer programs configured to perform the operations of the methods and encoded on computer storage devices. Details of one or more embodiments of the subject matter described herein are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, drawings, and claims. [Brief explanation of the drawings]

[0013] [Figure 1] 1 illustrates an exemplary particle crushing system. [Figure 2A] 1 illustrates an exemplary system for training a particle characterization model. [Figure 2B] 1 shows an overhead view of an exemplary sensor arrangement. [Figure 2C] 1 shows an overhead view of an exemplary sensor arrangement. [Figure 2D] 1 shows an overhead view of an exemplary sensor arrangement. [Figure 2E] FIG. 1 is a flow diagram showing a process for training and implementing a particle characterization model. [Figure 3A] 1 illustrates an exemplary system for training a predictive model to predict post-crush particle properties of crushed particles output from a crusher. [Figure 3B] FIG. 1 is a flow diagram illustrating a process for training a predictive model. [Figure 4A] 1 illustrates an exemplary system for optimizing crusher settings based on predicted post-crush characteristics. [Figure 4B] FIG. 1 is a flow diagram illustrating a process for optimizing crusher settings using predicted post-crush properties. [Figure 5A] 1 illustrates an exemplary system for optimizing crusher settings based on observed post-crush characteristics. [Figure 5B] FIG. 1 shows a flow diagram illustrating a process for optimizing crusher settings using observed post-crush properties. [Figure 6] 1 shows a schematic diagram of a computer system to which any of the computer-implemented methods and other techniques described herein may be applied.

[0014] Like reference numbers and designations in the various drawings indicate like elements. DETAILED DESCRIPTION OF THE INVENTION

[0015] FIG. 1 illustrates an exemplary rock crushing system 100. During operation, the crushing system 100 crushes aggregate particles 101 in a crusher 112. The crusher 112 can be any type of crusher, such as a cone jaw, roller, or impact. The crusher 112 can crush the aggregate particles 101 to a particular size and / or shape. The operation of the crusher 112 can be controlled by a control signal 126 from a control system 102. For example, the control signal 126 from the control system 102 can set a crusher setting 128 to cause the crusher 112 to increase or decrease the size of the crushed particles 105.

[0016] The control system 102 can analyze the crushed particles 105 using pre-crush sensor data 122 from the pre-crush sensors 104. The pre-crush sensors 104 generate sensor data from measurements of the particles before the particles are crushed by the crusher 112. The control system 102 is configured to control various aspects of the crushing process. For example, the control system 102 can store and execute one or more sets of computer instructions for controlling the execution of aspects of the crushing process described herein. The control system 102 can include a system of one or more computing devices. The computing devices can be, for example, a system of one or more servers. For example, a first server can be configured to receive and process data from the pre-crush sensors 104. Another server can be configured to interface with the crusher 112 and issue control commands based on the analysis results from the first server.

[0017] The pre-crush sensor 104 can include a variety of different sensors configured to measure various properties of the particles. For example, sensors used by the pre-crush sensor 104 can include, but are not limited to, optical sensors (e.g., visible light cameras, infrared cameras, near-infrared (NIR) sensors, dynamic optical microscope sensors) and mechanical sensors (e.g., sieves, sedigraphs, impact hammers, electrodynamic oscillators), as well as spectrometers. In some examples, diffuse reflectance spectroscopy across the visible, near-infrared, and short-wavelength-infrared spectral regions (400 nm to 2500 nm) can be used as a tool to assess particle strength.

[0018] Analysis of the aggregate particles 101 can be determined by, but is not limited to, NIR light detection and regression models to correlate images and reactant content in the sample. In some examples, sensor data from the pre-crush sensors 104 can be used to create a synthetic digital twin of the crushed particles 105.

[0019] The pre-crush sensor data 122 is used by the control system 102 to determine characteristics of the aggregate particles 101. For example, the particle characteristics may include, but are not limited to, particle size, shape, texture, surface area, sphericity, porosity, density, strength, and particle size distribution. In some examples, the pre-crush sensor data 122 may be used to determine the exposure of the particles to elements such as seawater.

[0020] The crushed particles 105 may be transported from the crusher 112 to a downstream processing system. In some examples, the crushed particles 105 may be transported by a series of augers and conveyors, such as conveyor 120. The crushed particles 105 pass a post-crush sensor 106 on their way to the stockpile 110. The post-crush sensor 106 generates sensor data from measurements of the particles after they are crushed by the crusher 112. In some examples, the post-crush sensor 106 is positioned vertically below the crusher 112. The post-crush sensor 106 may observe the crushed particles as they fall past the post-crush sensor 106 due to gravity.

[0021] The post-crush sensors 106 are configured to obtain post-crush sensor data 124 of the particles. For example, in some implementations, optical sensors may be arranged in an array along a conveyor or chute used to transport the particles. The post-crush sensors 106 may transmit the post-crush sensor data 124 to the control system 102. In some examples, the post-crush sensors 106 may include the same type of sensors as the pre-crush sensors 104. In some examples, the post-crush sensors 106 may have an arrangement that is the same as or similar to the arrangement of the pre-crush sensors 104.

[0022] Aggregate particles extracted from a quarry can have high variability in properties and quality. The aggregate particles can vary in mineral composition, size distribution, shape distribution, compressive strength, and / or specific gravity. The aggregate particles can vary in composition, for example, as a result of the specific geological formations in which they form, as a result of how they are generated, or both. The aggregate particles can be formed, for example, by excavation, dredging, or blasting. As a result, the particles input to the crushing equipment can vary significantly, leading to downstream variations in the output of crushed particles.

[0023] Parameters of aggregate particle crushing operations can vary. Crushing equipment can vary depending on equipment age, equipment wear, equipment compatibility with the aggregate input particles, operator skill, and equipment settings. Equipment settings can include the rate of aggregate addition, the distance of the mill cone or jaws from the mill wall, the speed of crusher operation, jaw opening, material feed rate, work surface opening size, and mill operating speed. As a result, variability in output crushed particles resulting from non-uniformity of the input particles is exacerbated by operational variability.

[0024] 2A illustrates an exemplary system 200 for training a particle characterization model 208. The system 200 includes a high-resolution sensor 206. The high-resolution sensor 206 is a high-data output sensor, or high-fidelity sensor, that generates high-resolution sensor data 224. The system 200 includes a low-resolution sensor 204. The low-resolution sensor 204 is a low-data output sensor, or low-fidelity sensor, that generates low-resolution sensor data 222.

[0025] The control system 102 stores a characterization model 208. The characterization model 208 includes a mapping 210 of high-resolution sensor data 224 to low-resolution sensor data 222. The characterization model 208 can be developed through a training process. In some examples, the training process can be performed in a laboratory setting.

[0026] Multiple sensors can be positioned to measure aggregate particles. In some examples, sensors can be positioned in an array or ring around the location where the aggregate particles fall into the stream, for example, off the conveyor belt. As shown in FIG. 2A , aggregate particles 201 fall downward due to gravity. The direction of gravity is represented by the z-axis of coordinate system 220. In the example of FIG. 2A , high-resolution sensor 206 is positioned vertically above low-resolution sensor 204. In some examples, high-resolution sensor 206 can have the same height as low-resolution sensor 204. In some examples, high-resolution sensor 206 can be positioned vertically below low-resolution sensor 204. In some examples, individual low-resolution sensors 204 can have different elevations from one another. In some examples, individual high-resolution sensors 206 can have different elevations from one another. Exemplary sensor placements are described in more detail with reference to FIGS. 2B , 2C , and 2D .

[0027] In some examples, the high resolution sensor 206, the low resolution sensor 204, or both, can measure the particles 201 while the particles are on the conveyor 120. For example, the high resolution sensor 206, the low resolution sensor 204, or both, can be positioned around the conveyor 120 and can measure the particles 201 while the particles are stationary or moving on the conveyor 120. In some examples, the low resolution sensor 204 can be positioned in a different location than the high resolution sensor 206. For example, the low resolution sensor 204 can be positioned around the location where the aggregate particles fall, and the high resolution sensor 206 can be positioned around or in a different location from the conveyor 120.

[0028] In some examples, the low-resolution sensor 204 can measure the aggregate particles 201 in a main flow of the aggregate particles 201, and the high-resolution sensor 206 can obtain measurements of a sample of the aggregate particles 201 from the main flow. For example, the low-resolution sensor 204 can be positioned to measure the main flow of the aggregate particles 201 on or falling from the conveyor 120, and the sample of the aggregate particles 201 can be diverted to a different sample measurement location. The high-resolution sensor 206 can obtain measurements of the sample at the sample measurement location. Each sample can contain, for example, several kilograms of aggregate particles. In some examples, a diversion gate can be used to divert the sample from the main flow. In this manner, the system 200 uses an integrated process for taking samples of aggregate particles for high-resolution scanning.

[0029] In some examples, the high-resolution sensors 206, the low-resolution sensors 204, or both can be arranged in a volumetric assembly. The low-resolution sensors 204, the high-resolution sensors 206, or both can be solar-powered and can incorporate wired or wireless communication systems for communicating with the control system 102. Data from multiple sensors can be aggregated to generate an estimate of the size of a particle passing through the sensor ring. For example, data from all of the low-resolution sensors 204 can be aggregated together, and data from all of the high-resolution sensors 206 can be aggregated together. Thus, sensor data can be generated using non-contact sensor measurements.

[0030] 2B, 2C, and 2D show overhead views of exemplary sensor arrangements. In some examples, high-resolution sensors can be interspersed with the same or similar number and / or spacing of low-resolution sensors. For example, referring to FIG. 2B, high-resolution sensors 206b, shaded in black, are interspersed with low-resolution sensors 204b, shaded in white, to form a ring or circle. The high-resolution sensors 206b and low-resolution sensors 204b can generate sensor data from measurements of aggregate particles falling through the ring, e.g., falling downward in the z-direction due to gravity. The high-resolution sensors 206b are offset from each other by approximately 90 degrees in the x-y plane. The low-resolution sensors 204b are offset from each other by approximately 90 degrees in the x-y plane. In some examples, each of the high-resolution sensors 206b is positioned at the same or approximately the same height along the z-axis. In some examples, the high-resolution sensors 206b and the low-resolution sensors 204b are positioned at the same or approximately the same height along the z-axis.

[0031] 2C, high-resolution sensors 206c, shaded in black, are interspersed with low-resolution sensors 204c, shaded in white, to form a ring. The high-resolution sensors 206c are offset from one another by approximately 120 degrees in the xy plane. The low-resolution sensors 204c are offset from one another by approximately 120 degrees in the xy plane.

[0032] In some examples, the ring of low-resolution sensors 204 can be positioned above or below the ring of high-resolution sensors 206. In some examples, the low-resolution sensors 204 can be offset from the high-resolution sensors 206 in the x-y plane. In some examples, each low-resolution sensor can be aligned vertically, e.g., in the z-direction, with one of the high-resolution sensors. Vertically aligned sensors can have the same or similar viewpoint in the x-y plane. In some examples, vertically aligned sensors can have the same or similar orientation or pose.

[0033] 2D , the low-resolution sensors 204d, shaded in white, are positioned vertically above the high-resolution sensors 206d, shaded in black. Each low-resolution sensor 204d is aligned or nearly aligned in the z-direction with one of the high-resolution sensors 206d. The fields of view of the low-resolution sensors 204d and the aligned high-resolution sensors 206d can overlap and can be the same or similar.

[0034] The high-resolution sensor 206 is used to acquire high-resolution sensor data 224 generated from measurements of the batch of aggregate particles 201 at the individual particle scale. The high-resolution sensor 206 can include, for example, a high-resolution laser displacement scanner, a stereo camera assembly, and a LiDAR sensor. The high-resolution sensor 206 can be, for example, a sensor with a spatial resolution of 1 millimeter or less (e.g., 500 microns or less, 100 microns or less). The high-resolution sensor 206 can generate data representing geometric and chemical properties of the aggregate particles. The properties can include, for example, the size, shape, and surface area of ​​the individual particles. The high-resolution sensor data 224 can include measurements of surface area measured at the micron scale.

[0035] The control system 102 can use the low-throughput, high-resolution sensor data 224 to determine properties of individual aggregate particles. The properties can include geometric properties, such as size, shape, texture, porosity, and surface area data for each individual particle in the batch of aggregate particles 201. The properties can include chemical properties, such as hydration, chemical composition, and oxidation state. The chemical composition, including mineral type, can indicate the crystalline structure and mechanical properties of the particle. The chemical composition can also indicate properties such as potential reactivity. The properties of each individual particle in the batch of aggregate particles 201 can be used to determine properties of the entire batch. For example, the control system 102 can determine a statistical representation of each property for the batch of aggregate particles 201.

[0036] The low-resolution sensor 204 can be used to acquire low-resolution sensor data 222 generated from measurements of batches of aggregate particles 201 at a batch scale. The low-resolution sensor 204 can include, for example, an ultrasonic sensor, a depth camera, a multi-camera array, a line scanner, and a monochrome camera. The low-resolution sensor 204 can be, for example, a sensor having a spatial resolution of one millimeter or greater (e.g., several millimeters or greater, one centimeter or greater, one inch or greater). The low-resolution sensor data 222 can include image data and / or pixel data generated from measurements of batches of aggregate particles 201 by the low-resolution sensor 204.

[0037] The control system 102 can store sensor position data for sensors, including the low-resolution sensors 204 and the high-resolution sensors 206. The sensor position data 211 can include, for each sensor, the height or altitude of the sensor. The sensor position data 211 can include, for each sensor, the position relative to the center of the stream of aggregate particles 201, the position relative to other sensors, or both. In some examples, the sensor position data 211 can include a two-dimensional coordinate position or a three-dimensional coordinate position. The coordinate position can be relative to a reference location, for example, the location where the aggregate particles 201 drop off the conveyor.

[0038] In some examples, the high-resolution sensor data 224 and the low-resolution sensor data 222 include, for each instance of the data, metadata indicating the position of the sensor that generated the instance of the data. In some examples, the sensor position data 211 includes an identifier for each sensor. The identifier may be associated with the location, orientation, perspective, pose, or any of these of the sensor. The high-resolution sensor data 224 and the low-resolution sensor data 222 may include, for each instance of the data, metadata including an identifier of the sensor that generated the instance of the data.

[0039] The control system 102 can aggregate data from the high-resolution and low-resolution sensors, as well as data indicating the location of each sensor relative to the aggregate stream and other sensors. The data can be aggregated to obtain an estimate of the size of an aggregate particle 201 passing through the center of the sensor ring.

[0040] In some examples, the low-resolution sensor data 222, the high-resolution sensor data 224, or both can be time-stamped. The characterization model can map the high-resolution sensor data 224 generated by a particular high-resolution sensor to the low-resolution sensor data 222 generated by a particular low-resolution sensor for the same particle based on the associated timestamps and based on the relative position of the particular high-resolution sensor with respect to the particular low-resolution sensor.

[0041] The characterization model 208 is trained using input data including high-resolution sensor data 224 and low-resolution sensor data 222. An exemplary characterization model 208 may be a neural network machine learning model. During the training process, the characterization model 208 receives as input the low-resolution sensor data 222 and the high-resolution sensor data 224. The training process uses the high-resolution sensor data as training data (e.g., a representation of approximate ground truth data). During the training process, the characterization model 208 determines correlations and / or mappings between the low-resolution sensor data 222 and the high-resolution sensor data 224 that represent the same batch of aggregate particles 201.

[0042] In some examples, the high-resolution sensor data 224 includes measurements of particle texture, particle surface area, or both. For a given shape and size of a particle, texture can be related to surface area. For example, a rough or jagged particle generally has a larger surface area than a smooth particle of the same size and shape.

[0043] The high-resolution sensor data 224 can include shape data that indicates the shape of the particle. The shape data can include, for example, data that indicates aspect ratio, convexity, concavity, or any combination thereof. The shape data can be used to determine the surface area of ​​the particle. For example, particles with a non-convex shape generally have a larger surface area per volume compared to particles with a convex shape.

[0044] The characterization model 208 includes mapping data 210 that indicates a correlation between the low-resolution sensor data 222 and the high-resolution sensor data 224. For example, the mapping data 210 can include a mapping between particle surface areas determined from the high-resolution sensor data 224 and particle shapes determined from the low-resolution sensor data 222.

[0045] In some examples, mapping data 210 can include correlations between two-dimensional pixel data generated by low-resolution sensor 204 and three-dimensional or four-dimensional pixel data generated by high-resolution sensor 206 for the same particle. Thus, characterization model 208 can be trained to approximate aggregate particle dimensions using lower-quality, lower-resolution sensor data. In other words, characterization model 208 is trained to characterize particles using only lower-quality (and lower-bandwidth) sensor data while approximating the accuracy achieved using higher-quality (and higher-bandwidth) sensor data. Characterization model 208 can apply a model of particle texture generated from high-resolution sensor data 224 to particle shape features determined using low-resolution sensor data 222. Thus, the trained characterization model 208 enables more computationally efficient and faster system operation for characterizing aggregate particles.

[0046] After the training process, the characterization model 208 can be periodically updated by scanning batches of aggregate particles at a laboratory scale, for example, using a high-resolution sensor such as a laser displacement scanner. At a laboratory scale, low volumes or low throughput aggregate particles can be analyzed. In some examples, a first volume of aggregate particles can be scanned for a first time period, after which the first volume of aggregate particles can be replaced with a second volume of aggregate particles. The second volume of aggregate particles can then be scanned for a second time period.

[0047] The low volume aggregate particles can include low mass aggregate particles. In some examples, the low mass aggregate particles are aggregate particles of a metric ton or less. In some examples, the throughput of aggregate particles used to train the characterization model 208 can be 100 kilograms / hour or less (e.g., 50 kilograms / hour or less, 10 kilograms / hour or less).

[0048] 2E is a flow diagram illustrating a process 250 for training and implementing a particle characterization model. Process 250 may be performed by one or more computing devices. For example, process 250 may be performed by control system 102. The operations of process 250 are described as being performed by the control system. However, some or all of the operations may be performed by various operational modules of the particle crushing system.

[0049] Process 250 includes acquiring 212 low-fidelity sensor data generated from measurements of the first portion of particles. For example, control system 102 can acquire low-fidelity sensor data or low-resolution sensor data 222 generated from measurements of the first portion of aggregate particles 201 by low-resolution sensor 204. In some examples, the low-fidelity sensor data is indicative of shape characteristics, size characteristics, or both of each particle in the first portion of particles.

[0050] Process 250 includes acquiring 214 high-fidelity sensor data generated from measurements of the first portion of particles. For example, control system 102 can acquire 214 high-fidelity sensor data generated from measurements of the first portion of aggregate particles 201 by high-resolution sensor 206 or high-resolution sensor data 224. In some examples, the high-fidelity sensor data indicates the surface area, texture, or both of each particle in the first portion of particles.

[0051] Process 250 includes training 216 a characterization model using the low-fidelity sensor data and the high-fidelity sensor data. For example, characterization model 208 can be trained using low-resolution sensor data 222 and high-resolution sensor data 224. In some examples, characterization model 208 can be trained by providing the high-fidelity sensor data as training data to the characterization model, processing the high-fidelity sensor data with the characterization model, and correlating the high-fidelity sensor data with the low-fidelity sensor data. In some examples, correlating the high-fidelity sensor data with the low-fidelity sensor data includes mapping surface area and / or texture characteristics indicated by the high-fidelity sensor data to shape and / or size characteristics indicated by the low-fidelity sensor data. In some examples, characterization model 208 includes a mapping between particle shape and surface area. In some examples, characterization model 208 includes a mapping between particle shape and texture. In some examples, characterization model 208 can be trained using a small number of aggregate particles. A small amount of aggregate particles can be, for example, aggregate particles having a total mass of 100 kilograms or less.

[0052] The process 250 includes determining 218 characteristics of the second portion of the particles using the characterization model. For example, the characterization model 208 may be used to determine the properties of the second portion of the particles. The trained characterization model 208 may determine the properties of the aggregate particles using low-resolution sensor data generated by a low-fidelity, high-throughput sensor. In some examples, the low-resolution sensor data may indicate shape and / or size characteristics of the second portion of the particles. The characterization model 208 may output the properties of the second portion of the particles, including data indicating the surface area and / or texture of the second portion of the particles.

[0053] To determine aggregate particle characteristics, the trained feature evaluation module 208 can apply texture and / or surface area models generated at low throughput during the training process to particle shape and / or size determined using high-throughput, low-fidelity sensor data. Thus, the characterization model 208 can determine aggregate particle characteristics without high-resolution sensor data acquired using high-fidelity, low-throughput sensors. The ability to characterize particles at high throughput improves the accuracy and efficiency of aggregate processing operations performed on an industrial scale. High throughput is a mass flow rate of aggregate particles that is greater than the mass flow rate of aggregate particles at low throughput. In some examples, the mass flow rate of aggregate particles at high throughput is at least 100 times the mass flow rate of aggregate particles at low throughput. In some examples, the mass flow rate of aggregate particles at high throughput is at least 1000 times the mass flow rate of aggregate particles at low throughput.

[0054] The trained characterization model 208 can be used to characterize pre-crushed particles, post-crushed particles, or both. For example, the characterization model 208 can receive as input sensor data generated from measurements of crushed or uncrushed particles and generate an output including properties of the crushed or uncrushed particles. The characterization model 208 can be used to characterize aggregate particles at a site, such as a quarry where large volumes of aggregate particles are produced. The second portion of particles can include a larger volume of aggregate particles compared to the first portion of particles. For example, the second portion of particles can be one metric ton or more of particles. In some examples, the mass of the second portion of particles is at least 100 times greater than the mass of the first portion of particles. In some examples, the mass of the second portion of particles is at least 1000 times greater than the mass of the first portion of particles. In some examples, the mass of the second portion of particles is at least 10,000 times greater than the mass of the first portion of particles.

[0055] 3A shows an exemplary system 300 for training a predictive model 310 to predict post-crush particle properties of crushed particles 105 output from a crusher 112. Generally, pre-crush sensor data 122 is generated by a low-resolution pre-crush sensor 104 from measurements of a batch of aggregate particles 101 before crushing the batch of aggregate particles 101 in the crusher 112. A characterization model 208 determines pre-crush properties 302 from the pre-crush sensor data 122. The predictive model 310 outputs predicted post-crush properties 312.

[0056] Post-crush sensor data 124 is generated by post-crush sensors 106 from measurements of the crushed particles output from the crusher 112. The characterization model 208 determines post-crush properties 304 from the post-crush sensor data 124. The post-crush properties may include geometric and chemical properties of the crushed particles 105. The post-crush properties 304 may be used as ground truth data for training a predictive model 310.

[0057] The evaluator 307 can compare the predicted post-crush characteristics 312 to the post-crush characteristics 304 to determine an error. The tuner 314 can adjust the parameters 316 of the predictive model 310 based on the error 308. Thus, the predictive model 310 can be trained over time to reduce the error between the predicted post-crush characteristics 312 and the post-crush characteristics 304 determined by the characterization model 208.

[0058] In some implementations, the control system 102 includes a set of operational modules for controlling different aspects of the fracturing process. The operational modules may be provided as one or more computer-executable software modules, hardware modules, or a combination thereof. For example, one or more of the operational modules may be implemented as blocks of software code having instructions that cause one or more processors of the control system 102 to perform the operations described herein. Additionally or alternatively, one or more of the operational modules may be implemented in an electronic circuit, such as, for example, a programmable logic circuit, a field programmable logic array (FPGA), or an application specific integrated circuit (ASIC). The operational modules may include a characterization model 208, a predictive model 310, an evaluator 307, and a tuner 314.

[0059] In some implementations, the predictive model 310 can include a machine learning model for estimating post-crush properties from measured pre-crush properties. The predictive model 310 can be, for example, a deterministic model such as a neural network or a probabilistic model such as a Gaussian process. In some examples, the machine learning model is trained on experimental data to receive pre-crush properties as input and generate a predicted output, e.g., an estimated post-crush property.

[0060] In some implementations, the machine learning model is a deep learning model that uses multiple layers of the model to generate an output for a received input. A deep neural network is a deep machine learning model that includes an output layer and one or more hidden layers, each applying a nonlinear transformation to a received input to generate an output. In some examples, the neural network may be a recurrent neural network. A recurrent neural network is a neural network that receives an input sequence and generates an output sequence from the input sequence. In particular, a recurrent neural network generates an output from a current input in the input sequence using some or all of the internal state of the network after processing a previous input in the input sequence. In some other implementations, the machine learning model is a convolutional neural network. In some implementations, the machine learning model is an ensemble of models that may include all or a subset of the architectures described above.

[0061] In some implementations, the machine learning model may be a feedforward autoencoder neural network. For example, the machine learning model may be a three-layer autoencoder neural network. The machine learning model may include an input layer, a hidden layer, and an output layer. In some implementations, the neural network does not have recurrent connections between layers. Each layer of the neural network may be fully connected to the next layer, and there may be no pruning between layers. The neural network may include an ADAM optimizer or any other multidimensional optimizer to train the network and calculate updated layer weights. In some implementations, the neural network may apply a mathematical transform, such as a convolutional transform, to input data before feeding the input data to the network.

[0062] In some implementations, the machine learning model may be supervised. For example, for each input provided to the model during training, the machine learning model may be instructed on what the correct output should be. The machine learning model may use batch training, training on a subset of examples before each adjustment instead of the entire set of available examples. This may improve the efficiency of training the model and may improve the generalizability of the model. The machine learning model may use fold cross-validation. For example, a portion of the data available for training (a "fold") may be excluded from training and used in a later testing phase to see how well the model generalizes. In some implementations, the machine learning model may be unsupervised. For example, the model may adjust itself based on the mathematical distance between examples rather than based on feedback about its performance.

[0063] A machine learning model can be trained to estimate the post-crush properties of particles based on the measured properties of the particles output from the crusher 112. In some examples, the machine learning model can be trained with experimentally determined data relating known pre-crush properties of particles to experimentally determined post-crush properties.

[0064] The pre-crush sensor data 122 can be generated from measurements of a large volume of aggregate particles. In some examples, a high volume or throughput of aggregate particles can include a high mass of aggregate particles. A high mass of aggregate particles can be, for example, one metric ton or more of aggregate particles. In some examples, a high throughput of aggregate particles can be a mass flow rate of aggregate particles of several metric tons per hour or more (e.g., 10 metric tons per hour or more, 20 metric tons per hour or more, 30 metric tons per hour or more). In some examples, the pre-crush sensor data 122 can be captured on-site or in a field setting, such as a quarry. In some examples, the pre-crush sensor data 122 can be captured continuously or continuously during a crushing operation. The high throughput of aggregate particles can be a mass flow rate that is greater than the mass flow rate of aggregate particles at a low throughput. For example, the high throughput of aggregate particles can be at least 100 times the mass flow rate of aggregate particles at a low throughput, or at least 1000 times the mass flow rate of aggregate particles at a low throughput.

[0065] The pre-crush sensors 104 can be positioned to measure the aggregate particles 101 before they undergo a crushing process by the crusher 112. For example, the pre-crush sensors 104 can be arranged in a ring around the location where the aggregate particles 101 fall into the stream, for example, off a conveyor belt. In some examples, the pre-crush sensors 104 are the same sensors as the low-resolution sensors 204 used to train the characterization model 208. In some examples, the pre-crush sensors 104 are the same type of sensors as the low-resolution sensors 204. In some examples, the pre-crush sensors have the same geometry as the low-resolution sensors 204. In some examples, the pre-crush sensors have a different geometry than the low-resolution sensors 204. The characterization model 208 can aggregate the pre-crush sensor data 122 from the pre-crush sensors 104 to generate estimates of the geometric and chemical characteristics of the particles passing through the sensor ring.

[0066] The trained characterization model 208 detects geometric and chemical properties using low-resolution, high-throughput characteristics in the field. For example, input low-resolution pre-crush sensor data 122 acquired by a set of low-resolution pre-crush sensors 104 from measurements of a batch of aggregate particles 101 in a manufacturing environment may be provided to the characterization model 208.

[0067] In some examples, the low-resolution pre-crush sensor data 122 is obtained at intervals to sample batches of particles. In one example, aggregate particles falling from a conveyor may be sampled by the pre-crush sensor 104 at one-hour intervals for a one-minute period for each sample. In another example, the aggregate particles may be sampled by the pre-crush sensor 104 at one-day intervals for a one-hour period for each sample.

[0068] The characterization model 208 outputs pre-crush properties 302 of the batch of aggregate particles 101. In some examples, the characterization model 208 uses a mapping 210 of high-resolution data to low-resolution data to determine the pre-crush properties 302. The pre-crush properties 302 can include geometric properties, chemical properties, or both. In some examples, the output data includes estimated geometric properties of each particle in the batch of aggregate particles. In some examples, the output data includes averaged geometric properties of the batch of aggregate particles. In some examples, the output data indicates a distribution of particle properties, e.g., data representing a histogram distribution curve for each of a plurality of properties.

[0069] The chemical composition, size, and shape of a particle affect the crushing behavior of the particle when crushed. Thus, crushing particles with different properties may result in crushed particles with different properties. The predictive model 310 may be trained to predict the post-crush properties of particles based on their respective pre-crush properties.

[0070] Inputs to the predictive model 310 may include, for example, pre-crushing properties 302 generated by the characterization model 208. For example, the predictive model may receive as input pre-crushing properties 302 that indicate the geometric and chemical properties of particles 101 to be crushed by a rock crushing system, e.g., a crusher 112. In some examples, the pre-crushing properties 302 may indicate rock quality, e.g., shape characteristics, roughness, aspect ratio, volume.

[0071] The predictive model 310 may receive as input data indicative of rock crushing system measurements and settings. In some examples, the predictive model 310 may receive as input crusher measurements, such as the displacement rate and force required to achieve a particular displacement amount when crushing particles 101. In some examples, the predictive model 310 may receive as input crusher settings 128 representing the operating parameters of the crusher 112, such as material feed rate, work face opening size, and crusher operating speed, crushing pressure, and crusher lubricant flow rate.

[0072] The predictive model 310 can process input data to generate output data including predicted properties of the particles after crushing. In some examples, the predictive model 310 can be trained to predict or simulate the behavior of the crusher 112 when crushing the particles 101.

[0073] The predictive model 310 can determine a likely product output, such as, for example, a predicted post-crush property. Parameters of the predictive model can be adjusted based on comparing the predicted post-crush property 312 to the measured property of the particle after crushing. In some examples, the measured property of the particle is determined by providing low-resolution post-crush sensor data 124 to the characterization model 208. After training, the predictive model can be used to predict the post-crush property of the particle.

[0074] In some examples, the control system 102 can generate a data set from multiple batches of particles over time. tm:i-1(xy), and D is the time interval t i is a dataset for input x and output y in a given time interval. For a given time interval, the predictive model 310 can be trained using the corresponding dataset D. The output crushed particle properties are i The output of the predictive model 310 can be expressed using Equation 1:

[0075]

number

[0076] The predictive model 310 predicts the properties of aggregate particles exiting the crusher 112 as a function of the inputs, where m represents the number of past time intervals included in the predictive model 310. The predictive model 310, trained on the moving horizon dataset, can then be used to predict the performance of the rock crusher 112 as a function of the input parameters.

[0077] The pre-crush sensor data 122 and post-crush sensor data 124 may be time-stamped and synchronized with the flow rate of particles through the crushing system. For example, based on the crusher settings 128, the control system 102 may determine an estimated time for an individual particle 101 to travel from the location of the pre-crush sensor 104 to the crusher 112, and for the resulting crushed particle to travel from the crusher 112 to the location of the post-crush sensor 116. Based on the estimated travel time and the time-stamped sensor data, the evaluator 307 may compare the post-crush signature 302 to the pre-crush signature 302 for the same particle or set of particles.

[0078] In one example, the estimated travel time between the pre-crush sensor 104 and the post-crush sensor 106 may be 30 seconds. The characterization model 208 may determine a pre-crush characteristic 302 of a first particle 101 passing the pre-crush sensor 104 at 10:00:00 AM based on pre-crush sensor data time-stamped at 10:00:00 AM. The predictive model 310 may determine a predicted post-crush characteristic for the first particle. The characterization model 208 may then determine a post-crush characteristic 304 of a broken particle 105 passing the post-crush sensor 106 30 seconds later based on post-crush sensor data time-stamped at 10:00:30 AM. The evaluator 307 may compare the predicted post-crush characteristic 312 to the post-crush characteristic 304 of the first particle to determine an error 308. In some examples, the evaluator 307 may compare the predicted post-crash characteristics 312 to the post-crash characteristics 304 of data captured within a time window around the predicted travel time. For example, the evaluator 107 may compare the predicted post-crash characteristics 312 to the post-crash characteristics 304 averaged over a time window of ±1 second from the expected travel time, e.g., from 10:00:29 AM to 10:00:31 AM.

[0079] 3B is a flow diagram illustrating a process 350 for training the predictive model 310. The process 350 may be performed by one or more computing devices. For example, the process 350 may be performed by the control system 102. The operations of the process 350 are described as being performed by the control system. However, some or all of the operations may be performed by various operational modules of the particle crushing system.

[0080] Process 350 includes obtaining pre-crushed particle data indicative of some characteristics of the particles. (332) For example, predictive model 310 receives as input pre-crushed characteristics 302 determined by characterization model 208 through measurements of particles 101 being crushed by crusher 112.

[0081] The process 350 includes obtaining 334 configuration data indicative of the configuration of the crushing system. For example, the predictive model 310 receives as input the crusher configuration 128 indicative of the configuration of the crusher 112.

[0082] The process 350 includes obtaining (336) post-crush particle data indicative of particle characteristics of the portion of the particles after crushing the portion of the particles using the crushing system. For example, the characterization model 208 generates post-crush characteristics 304 from post-crush sensor data 124 generated through measurements of the crushed particles 105 output by the crusher 112. An evaluator 307 compares the post-crush characteristics 304 with predicted post-crush characteristics 312 determined by a predictive model 310. An adjuster 314 adjusts parameters 316 of the predictive model 310 based on an error 308 between the post-crush characteristics 304 and the predicted post-crush characteristics 312.

[0083] The process 350 includes training a model to predict post-crush properties using the pre-crush particle data, the configuration data, and the post-crush particle data (338).

[0084] In general, the operating parameters and settings of the crusher 112 can be adjusted based on the measured or predicted characteristics of the crushed particles. Figure 4A shows an example system 400 for optimizing the crusher settings 128 based on the predicted post-crush characteristics 312.

[0085] The predictive model 310 receives as input the pre-crush properties 302 determined by the characterization model 208 using the pre-crush sensor data 122. The predictive model 310 correlates the input geometry and chemistry to predicted mill output.

[0086] The predictive model 310 outputs predicted post-crush characteristics 312 to the optimization model 410. The optimization model 410 can be, for example, a machine learning model, a Gaussian model, or a hybrid model. In some examples, the optimization model 410 is a data-driven model or a physics-based model. The optimization model 410 can determine optimized crusher settings using, for example, a Bayesian algorithm, a genetic algorithm, active learning, Q-learning, or any combination thereof. In some examples, the optimization model 410 can include a PID control algorithm.

[0087] The optimization model 410 can output a control signal 126 to adjust the settings of the grinder 112. In some examples, the grinder 112 is controlled by a controller 420 that is separate from the control system 102. The controller 420 receives the control signal 126 from the control system 102 and adjusts the operation of the grinder 112 based on the control signal 126.

[0088] Crusher settings can include, for example, material feed rate, crushing rate, mass or volumetric flow rate of aggregate particles through the crusher 112, crusher work face opening size, or any of these. Crusher feed rate 112 can be measured by mass flow rate, e.g., in tons / hour. Crusher feed rate 112 is related to feed belt speed. Crusher crushing rate 112 can be measured as a cycle rate, e.g., in cycles per minute. A cycle can be, for example, the opening and closing of the jaws of a jaw crusher. In some examples, a cycle is a rotation in a gyratory crusher.

[0089] In some examples, the optimization model 410 can determine an error by comparing the predicted post-crush properties 312 to target properties 422. The target properties 422 can include a target size and shape distribution. In some examples, the target properties 422 can be input to the control system 102 by an input / output device, such as a computing device 440. The target properties 422 can be provided to the control system 102 as input by a user via the computing device 440.

[0090] In some examples, the target characteristic 422 can be generated by scanning a sample of particles having desired characteristics. For example, the sample of particles can be scanned using the post-crush sensor 106, and the post-crush sensor data 124 generated by the post-crush sensor can be provided to the characterization model 208. The characterization model 208 can determine the characteristics of the sample of particles. The characteristics of the sample of particles can be stored by the control system 102 and stored as the target characteristic 422.

[0091] Based on the error between the predicted post-crush characteristic 312 and the target characteristic 422, the optimization model 410 can output a control signal 126 to adjust the crusher settings 128. The predictive model 310 can use the updated crusher settings 128 to determine an updated predicted post-crush characteristic 312. In some examples, the control system 102 can iteratively update the crusher settings 128 and determine the predicted post-crush characteristic 312 until the error between the predicted post-crush characteristic 312 and the target characteristic 422 is equal to or less than a threshold error.

[0092] In this manner, the control system 102 can execute a feedforward control process to adjust the settings of the crusher 112 to more closely align the predicted post-crush characteristics 312 with the target characteristics 422. In some examples, the optimization model 410 can adjust the crusher settings 128 to minimize the error between the predicted post-crush characteristics 312 and the target characteristics 422, subject to crusher setting constraints.

[0093] The control system 102 can obtain post-crush sensor data 124 and generate updated control signals 126 over time. In some examples, the control system 102 can continuously or repeatedly update the control signals 126 until the predicted post-crush characteristics 312 match the target characteristics 422 within a threshold error.

[0094] 4B is a flow diagram illustrating a process 450 for optimizing crusher settings using predicted post-crush characteristics. Process 450 may be performed by one or more computing devices. For example, process 450 may be performed by control system 102. The operations of process 450 are described as being performed by the control system. However, some or all of the operations may be performed by various operational modules of the particle crushing system.

[0095] The process 450 includes obtaining 402 pre-crush particle data indicative of some characteristics of the particles prior to crushing the particles in the crushing system. For example, the predictive model 310 can obtain the pre-crush characteristics 302 generated by the characterization model 208 using the pre-crush sensor data 122. The pre-crush sensor data 122 is generated from measurements of a batch of particles 101 input to the crusher 112.

[0096] The process 450 includes processing the pre-crushed particle data using a predictive model to obtain 404 an output including predicted post-crushed properties of a portion of the particles. For example, the predictive model 310 may process the pre-crushed properties 302 and output predicted post-crushed properties 312. The predicted post-crushed properties 312 represented the predicted properties of the crushed particles 105 output by the crusher 112.

[0097] The process 450 includes adjusting 406 settings of the crushing system based on the error between the predicted post-crush characteristics and the target characteristics. For example, the optimization model 410 can output a control signal 126 that adjusts the crusher settings 128 based on the error between the predicted post-crush characteristics 312 and the target characteristics 422.

[0098] The estimated post-crushing properties 304 of the particles output by the characterization model 208 can be used to adjust parameters of the process for crushing the aggregate. FIG. 5A shows an example system 500 for optimizing crusher settings 128 based on the observed post-crushing properties 304. The control system 102 can determine an error by comparing the estimated geometric properties of the batch of crushed aggregate particles output by the crusher 112 to target properties. In some examples, the estimated properties can be determined using low-resolution sensor data processed by the characterization model 208.

[0099] The characterization model 208 receives as input the post-crush sensor data 124. The characterization model 208 outputs the post-crush characteristics 304 to an optimization model 410. The optimization model 410 can be trained to optimize the settings of the crusher 112 based on the aggregate characteristics being input to the crusher 112. The crusher settings 128 include, for example, the crusher material feed rate, the crusher operating speed, the crusher working face opening, the speed of the crusher operation, and other settings.

[0100] In some examples, the optimization model 410 can determine an error by comparing the post-crush characteristic 304 to a target characteristic 422. Based on the error between the post-crush characteristic 304 and the target characteristic 422, the optimization model 410 can output a control signal 126 to adjust the crusher settings 128. In some examples, the control system 102 outputs the control signal 126 to a controller 420 of the crusher 112. The controller 420 receives the control signal 126 from the control system 102 and adjusts the operation of the crusher 112 based on the control signal 126.

[0101] In some examples, the optimization model 410 can adjust the crusher settings 128 to minimize the error between the post-crush characteristic 304 and the target characteristic 422, subject to crusher setting constraints. In some examples, the control system 102 can iteratively update the crusher settings 128 until the error between the post-crush characteristic 304 and the target characteristic 422 is equal to or less than a threshold error. In this manner, the control system 102 can perform a feedback control process to adjust the settings of the crusher 112 to more closely align the post-crush characteristic 304 with the target characteristic 422.

[0102] The control system 102 can obtain post-crush sensor data 124 and generate updated control signals 126 over time. In some examples, the control system 102 can continuously or repeatedly update the control signals 126 until the post-crush characteristics 304 match the target characteristics 422 within a threshold error.

[0103] 5B is a flow diagram illustrating a process 550 for optimizing crusher settings using observed post-crushing characteristics. Process 550 can be performed by one or more computing devices. For example, process 550 can be performed by control system 102. The operations of process 550 are described as being performed by the control system. However, some or all of the operations may be performed by various operational modules of the particle crushing system.

[0104] The process 550 includes obtaining 502 sensor data generated from measurements of a portion of the crushed particles output by the crushing system. For example, the control system 102 obtains post-crush sensor data 124 generated from measurements of the crushed particles 505 output by the crusher 112.

[0105] The process 550 includes processing 504 the sensor data with a characterization model to obtain an output that includes some characteristics of the crushed particle. For example, the characterization model 208 can process the post-crush sensor data 124 to output post-crush characteristics 304 of the crushed particle 505.

[0106] The process 550 includes adjusting 506 settings of the crushing system based on the error between the predicted post-crush characteristic and the target characteristic. For example, the optimization model 410 can output a control signal 126 that adjusts the crusher setting 128 based on the error between the post-crush characteristic 304 and the target characteristic 422.

[0107] Systems 400 and 500 can be implemented to use the output crushed particle measurements to suggest changes to operational settings to optimize a desired objective function representing the equipment product output. Adjusting settings over time can compensate for aggregate particle property drift, crushing equipment parameter drift, or both. The real-time optimization process can minimize the error between the desired output (target characteristic 422) and the simulated output (predicted post-crushing characteristic 312), subject to constraints on the rock crusher settings. The real-time optimization process can minimize the error between the desired output (target characteristic 422) and the actual output (post-crushing characteristic 304), subject to constraints on the rock crusher settings. In some examples, the optimization model 410 can be trained to reduce and / or optimize both the error between the target characteristic 422 and the predicted post-crushing characteristic 312 and the error between the target characteristic 422 and the post-crushing characteristic 304.

[0108] FIG. 6 is a schematic diagram of a computer system 600. System 600, according to some implementations, can be used to perform the operations described in connection with any of the aforementioned computer-implemented methods. In some implementations, the computing systems and devices and functional operations described herein can be implemented in digital electronic circuitry, tangibly embodied computer software or firmware, computer hardware, including the structures disclosed herein (e.g., system 600) and their structural equivalents, or one or more combinations thereof. System 600 is intended to include various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers, including those installed on base or pod units of modular vehicles. System 600 can also include mobile devices, such as personal digital assistants, mobile phones, smartphones, and other similar computing devices. Furthermore, the system can include portable storage media, such as a Universal Serial Bus (USB) flash drive. For example, a USB flash drive may store an operating system and other applications. A USB flash drive may include input / output components such as a wireless transducer or a USB connector that may be inserted into a USB port on another computing device.

[0109] System 600 includes a processor 610, a memory 620, a storage device 630, and an input / output device 640. Each of the components 610, 620, 630, and 640 are interconnected using a system bus 650. Processor 610 is capable of processing instructions for execution within system 600. The processor may be designed using any of several architectures. For example, processor 610 may be a Complex Instruction Set Computer (CISC) processor, a Reduced Instruction Set Computer (RISC) processor, or a Minimal Instruction Set Computer (MISC) processor.

[0110] In one implementation, the processor 610 is a single-threaded processor. In another implementation, the processor 610 is a multi-threaded processor. The processor 610 can process instructions stored in the memory 620 or the storage device 630 to display graphical information for a user interface on the input / output device 640.

[0111] The memory 620 stores information within the system 600. In one implementation, the memory 620 is a non-transitory computer-readable medium. In one implementation, the memory 620 is a volatile memory unit. In another implementation, the memory 620 is a non-volatile memory unit.

[0112] The storage device 630 can provide mass storage for the system 600. In one implementation, the storage device 630 is a computer-readable medium. In various different implementations, the storage device 630 can be a floppy disk device, a hard disk device, an optical disk device, or a tape device.

[0113] The input / output device(s) 640 provide input / output operations for the system 600. In one implementation, the input / output device(s) 640 include a keyboard and / or a pointing device. In another implementation, the input / output device(s) 640 include a display unit for displaying a graphical user interface.

[0114] The described features can be implemented in digital electronic circuitry, or computer hardware, firmware, software, or combinations thereof. The apparatus can be implemented in a computer program product tangibly embodied in an information carrier, e.g., a machine-readable storage device, for execution by a programmable processor, and the method steps can be performed by the programmable processor executing a program of instructions that performs the functions of the described implementation by operating on input data and generating output. The described features can advantageously be implemented in one or more computer programs executable on a programmable system including at least one programmable processor coupled to receive data and instructions from a data storage system, at least one input device, and at least one output device, and to transmit data and instructions to the data storage system. A computer program is a set of instructions that can be used directly or indirectly in a computer to perform a particular activity or bring about a particular result. Computer programs can be written in any type of programming language, including compiled or interpreted languages, and can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0115] Processors suitable for executing a program of instructions include, by way of example, both general-purpose and special-purpose microprocessors, and the sole processor or one of multiple processors of any kind of computer. Generally, a processor will receive instructions and data from a read-only memory or a random-access memory or both. The essential elements of a computer are a processor for executing instructions and one or more memories for storing instructions and data. Generally, a computer also includes, or is operatively coupled to communicate with, one or more mass storage devices for storing data files; such devices include magnetic disks, such as internal hard disks and removable disks, magneto-optical disks, and optical disks. Suitable storage devices for tangibly embodying computer program instructions and data include, by way of example, semiconductor memory devices such as EPROM, EEPROM, and flash memory devices; magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and memory may be supplemented by, or incorporated in, application-specific integrated circuits (ASICs).

[0116] To provide for user interaction, this feature can be implemented on a computer having a display device, such as a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user, and a keyboard and pointing device, such as a mouse or trackball, by which the user can provide input to the computer. In addition, such activity can be implemented via touchscreen flat panel displays and other suitable mechanisms.

[0117] Such features may be implemented in a computer system that includes back-end components such as a data server, or includes middleware components such as an application server or an Internet server, or includes front-end components such as a client computer having a graphical user interface or an Internet browser, or any combination thereof. The components of the system may be connected by any form or medium of digital data communication, such as a communications network. Examples of communications networks include a local area network ("LAN"), a wide area network ("WAN"), a peer-to-peer network (with ad hoc or static members), a grid computing infrastructure, and the Internet.

[0118] A computer system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a network, such as that described. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0119] In addition to the embodiments described above, the following embodiments are also innovative.

[0120] Embodiment 1 is a method including: obtaining pre-crush particle data indicative of characteristics of a portion of the particles before crushing the portion of the particles with a crushing system; obtaining configuration data indicative of one or more settings of the crushing system; obtaining post-crush particle data indicative of characteristics of the portion of the particles after crushing the portion of the particles with the crushing system; training a predictive model using the pre-crush particle data, the configuration data, and the post-crush particle data, the training including processing the pre-crush particle data and the configuration data using the predictive model to obtain a corresponding output of the predictive model including predicted characteristics of the portion of the particles after crushing the portion of the particles with the crushing system; and adjusting parameters of the predictive model based on comparing the output of the predictive model to the post-crush particle data.

[0121] Embodiment 2 is the method of any of the preceding embodiments, wherein the pre-crushing particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of a portion of the particles prior to crushing the portion of the particles with a crushing system.

[0122] Embodiment 3 is the method of any of the preceding embodiments, wherein the post-crushing particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of a portion of the particles after crushing the portion of the particles with a crushing system.

[0123] Embodiment 4 is the method of any of the preceding embodiments, including using the trained predictive model to determine predicted post-crushing properties of the second portion of the particles by providing second pre-crushing particle data to the predictive model indicative of geometric and chemical properties of the second portion of the particles before crushing the second portion of the particles with the crushing system, and receiving the predicted properties of the second portion of the particles as output from the predictive model after crushing the second portion of the particles with the crushing system.

[0124] Embodiment 5 is the method of any of the previous embodiments, wherein the characteristics include geometric characteristics including at least one of size, shape, surface area, or sphericity.

[0125] Embodiment 6 is the method of any of the previous embodiments, wherein the properties indicated by the pre-crush particle data include chemical properties including at least one of chemical composition, mineral type, crystalline structure, or reactivity.

[0126] Embodiment 7 is the method of any of the previous embodiments, wherein the crushing system settings include at least one of material feed rate, work surface opening size, or crusher operating speed.

[0127] Example 8 is the method of any of the preceding examples, wherein the predictive model comprises a neural network model.

[0128] Embodiment 9 is a method for predicting post-crush properties of a portion of particles, the post-crush properties including properties of the portion of particles after crushing by a crushing system, where predicting includes obtaining pre-crush particle data indicating properties of the portion of particles before crushing the portion of particles with the crushing system, obtaining configuration data indicative of one or more settings of the crushing system, and processing the pre-crush particle data and the configuration data using a predictive model to obtain corresponding output of the predictive model, the corresponding output including the predicted post-crush properties of the portion of particles; determining an error between the predicted post-crush properties of the portion of particles and target properties of the crushed particles; and adjusting one or more settings of the crushing system based on the error between the predicted post-crush properties of the portion of particles and the target properties of the crushed particles.

[0129] Embodiment 10 is the method of any of the preceding embodiments, wherein the pre-crushing particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of a portion of the particles before crushing the portion of the particles with the crushing system.

[0130] Embodiment 11 is the method of any of the previous embodiments, wherein the settings of the crushing system include at least one of a material feed rate, a work surface opening size, or an operating speed.

[0131] Embodiment 12 is a method of any of the preceding embodiments, including iteratively predicting post-crush characteristics of the crushed particles output by the crushing system, and adjusting settings of the crushing system until an error between the predicted post-crush characteristics of the crushed particles and the target characteristics of the crushed particles is less than or equal to a threshold error.

[0132] Embodiment 13 is the method of any of the previous embodiments, wherein the post-crush properties include geometric properties including at least one of size, shape, surface area, or sphericity.

[0133] Embodiment 14 is a method of any of the preceding embodiments, including determining characteristics of a portion of the crushed particles output by the crushing system, determining an error between the characteristics of the portion of the crushed particles and target characteristics of the crushed particles, and adjusting one or more settings of the crushing system that outputs the portion of the crushed particles based on the error between the characteristics of the portion of the crushed particles and the target characteristics of the crushed particles.

[0134] Embodiment 15 is the method of any of the previous embodiments, wherein the characteristics include geometric characteristics including at least one of size, shape, surface area, or sphericity.

[0135] Embodiment 16 is a method of any of the preceding embodiments, in which determining the characteristics of a portion of the crushed particles output by the crushing system includes obtaining sensor data generated from measurements of the portion of the crushed particles by a set of sensors, and processing the sensor data using a characterization model to obtain the characteristics of the portion of the crushed particles.

[0136] Embodiment 17 is the method of any of the preceding embodiments, wherein the set of sensors includes at least one of an ultrasonic sensor, a depth camera, a multi-camera array, a monochrome camera, and a line scanner.

[0137] Example 18 is the method of any of the previous examples, wherein each of the sensors has a spatial resolution of 1 millimeter or greater.

[0138] Embodiment 19 is the method of any of the previous embodiments, wherein the one or more settings of the crushing system include at least one of a material feed rate, a work surface opening size, or an operating speed.

[0139] Embodiment 20 is a method of any of the preceding embodiments, including iteratively determining characteristics of the crushed particles output by the crushing system, and adjusting settings of the crushing system until an error between the characteristics of the crushed particles and the target characteristics of the crushed particles is less than or equal to a threshold error.

[0140] Embodiment 21 is a method of any of the preceding embodiments, wherein the pre-crushing particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of a portion of the particles before crushing the portion of the particles with a crushing system.

[0141] Embodiment 22 is the method of any of the preceding embodiments, wherein the post-crush particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of a portion of the particle after crushing the portion of the particle with a crushing system.

[0142] Embodiment 23 is a method of any of the preceding embodiments, including using the trained predictive model to determine predicted post-crushing properties of the second portion of the particles by providing second pre-crushing particle data to the predictive model indicating geometric and chemical properties of the second portion of the particles before crushing the second portion of the particles with the crushing system, and receiving predicted properties of the second portion of the particles as output from the predictive model after crushing the second portion of the particles with the crushing system.

[0143] Embodiment 24 is the method of any of the previous embodiments, wherein the characteristics include geometric characteristics including at least one of size, shape, surface area, or sphericity.

[0144] Embodiment 25 is the method of any of the preceding embodiments, wherein the properties indicated by the pre-crush particle data include chemical properties including at least one of chemical composition, mineral type, crystalline structure, or reactivity.

[0145] Embodiment 26 is the method of any of the preceding embodiments, wherein the settings of the crushing system include at least one of a material feed rate, a work surface opening size, or a crusher operating speed.

[0146] Example 27 is the method of any of the preceding examples, wherein the predictive model comprises a neural network model.

[0147] While this specification contains details of many specific implementations, these should not be construed as limitations on the scope of any invention or what may be claimed, but rather as descriptions of features specific to particular implementations of a particular invention. Certain features described herein in the context of separate implementations may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented in multiple implementations separately or in any suitable subcombination. Furthermore, while features may be described above as working in a particular combination, and may even initially be claimed as such, one or more features from a claimed combination may, in some cases, be deleted from the combination, and the claimed combination may be directed to a subcombination or a variation of the subcombination.

[0148] Similarly, although operations are depicted in the figures in a particular order, this should not be understood as requiring such operations to be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desired results. In certain situations, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system components in the above-described implementations should not be understood as requiring such separation in all implementations, and it should be understood that the described program components and systems may generally be integrated together in a single software product or packaged in multiple software products.

[0149] Thus, particular implementations of the present subject matter have been described. Other implementations are within the scope of the following claims. In some instances, the actions recited in the claims can be performed in a different order and still achieve desirable results. Moreover, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In certain implementations, multitasking and parallel processing may be advantageous.

[0150] While this specification contains many details of specific implementations, these should not be construed as limitations on the scope of what is claimed, which is defined by the claims themselves, but rather as descriptions of features that may be specific to particular embodiments of a particular invention. Certain features described herein in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Furthermore, while features may be described above as working in a particular combination and initially claimed as such, one or more features from a claimed combination can, in some cases, be deleted from the combination, and the claims may be directed to the subcombination or variations of the subcombination.

[0151] As used herein, the term "real-time" can refer to transmitting or processing data without intentional delay, taking into account the processing limitations of the system, the time required to accurately acquire the data, and the rate of change of the data. While some actual delay may occur, the delay is generally imperceptible to a user. The term "real-time" can refer to performing an action without intentional delay, taking into account the processing limitations of the system, the time required to accurately analyze the data, and the time required to generate a control signal to perform the action. Real-time performance of an action can include a delay of 10 seconds or less (e.g., a delay of 5 seconds or less, 3 seconds or less, 1 second or less) between the detection of a condition and the initiation of the action.

[0152] Specific embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results. As one example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some instances, multitasking and parallel processing may be advantageous.

Claims

1. 1. A method comprising: obtaining pre-crush particle data characteristic of the portion of the particles prior to crushing the portion of the particles using a crushing system; obtaining configuration data indicative of one or more settings of the fracturing system; obtaining post-crush particle data characteristic of the portion of the particle after crushing the portion of the particle using the crushing system; training a predictive model using the pre-crush particle data, the configuration data, and the post-crush particle data; The training processing the pre-crushing particle data and the configuration data using the predictive model to obtain a corresponding output of the predictive model, the corresponding output including a predicted property of the portion of the particles after crushing the portion of the particles in the crushing system; and adjusting parameters of the predictive model based on comparing the output of the predictive model to the post-crush particle data.

2. 2. The method of claim 1, wherein the pre-crush particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of the portion of particles prior to crushing the portion of particles in the crushing system.

3. 2. The method of claim 1, wherein the post-crush particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of the portion of the particle after crushing the portion of the particle with the crushing system.

4. determining predicted post-crush properties of a second portion of particles using the trained predictive model; providing second pre-crush particle data to the predictive model indicative of geometric and chemical characteristics of the second portion of the particles prior to crushing the second portion of the particles with the crushing system; and receiving as output from the predictive model predicted properties of the second portion of the particles after crushing the second portion of the particles using the crushing system.

5. The method of claim 1 , wherein the characteristics include geometric characteristics including at least one of size, shape, surface area, or sphericity.

6. The method of claim 1 , wherein the properties indicated by the pre-crushed particle data include chemical properties including at least one of chemical composition, mineral type, crystalline structure, or reactivity.

7. The method of claim 1 , wherein the settings of the crushing system include at least one of a material feed rate, a work surface opening size, or a crusher operating speed.

8. The method of claim 1 , wherein the predictive model comprises a neural network model.

9. 1. A method comprising: Predicting post-crush properties of a portion of a particle, the post-crush properties including properties of the portion of the particle after crushing by a crushing system, and predicting the post-crush properties of the portion of the particle includes: obtaining pre-crush particle data characteristic of the portion of the particles prior to crushing the portion of the particles using a crushing system; obtaining configuration data indicative of one or more settings of the fracturing system; processing the pre-crush particle data and the configuration data using a predictive model to obtain corresponding outputs of the predictive model, the corresponding outputs including predicted post-crush properties of a portion of the particles; predicting post-crush properties of a portion of the particle, including: determining an error between the predicted crushed properties of a portion of the particles and a target property of the crushed particles; and adjusting one or more settings of the crushing system based on the error between the predicted post-crush properties of the portion of the particles and target properties of the crushed particles.

10. 10. The method of claim 9, wherein the pre-crush particle data is generated by a characterization model configured to determine particle characteristics from sensor data generated from measurements of the portion of particles prior to crushing the portion of particles in the crushing system.

11. The method of claim 9 , wherein the settings of the crushing system include at least one of a material feed rate, a work surface opening size, or an operating speed.

12. iteratively predicting post-crush properties of crushed particles output by the crushing system and adjusting settings of the crushing system until an error between the predicted post-crush properties of the crushed particles and a target property of the crushed particles is less than or equal to a threshold error; 10. The method of claim 9, comprising:

13. The method of claim 9 , wherein the post-crush properties include geometric properties including at least one of size, shape, surface area, or sphericity.

14. 1. A method comprising: determining a characteristic of a portion of the fragmented particles output by the fragmentation system; determining an error between the characteristic of the portion of the broken particles and a target characteristic of the broken particles; and adjusting one or more settings of the fragmentation system that outputs the portion of the fragmented particles based on the error between the characteristic of the portion of the fragmented particles and the target characteristic of the fragmented particles.

15. The method of claim 14 , wherein the characteristics include geometric characteristics including at least one of size, shape, surface area, or sphericity.

16. determining the characteristics of a portion of the broken particles output by the break-down system; acquiring sensor data generated from measurements of a portion of the broken down particles by a set of sensors; and processing the sensor data with a characterization model to obtain the characteristics of the portion of the broken particle.

17. The method of claim 16 , wherein the set of sensors includes at least one of an ultrasonic sensor, a depth camera, a multi-camera array, a monochrome camera, and a line scanner.

18. The method of claim 16 , wherein each of the sensors has a spatial resolution of 1 millimeter or better.

19. 15. The method of claim 14, wherein the one or more settings of the crushing system include at least one of a material feed rate, a work surface opening size, or an operating speed.

20. iteratively determining a characteristic of the broken particle output by the break-down system and adjusting a setting of the break-down system until the error between the characteristic of the broken particle and the target characteristic of the broken particle is equal to or less than a threshold error; 15. The method of claim 14, comprising:

Images