End-to-end platform for managing circulation economy of waste

Abstract

To provide a technology for managing reuse of molecular components incorporated in waste material.SOLUTION: A system and a method for managing a chemical recycling process include accessing characterization data of a raw material, the characterization data including one or more spectra collected by one or more spectroscopic methods. This method includes using the characterization data to predict a set of constituent materials included in the raw material. This method includes using the predicted set of constituent materials to predict a material composition of the raw material. This method includes using the predicted material composition of the raw material, at least in part, to identify one or more target products. This method includes generating a set of chemical reaction schemes that enable conversion of at least part of the raw material into one or more target products. This method also includes storing the material composition of the raw material, the one or more target products, and identification of the set of chemical reaction schemes in a data store.SELECTED DRAWING: Figure 5

Description

[Background technology]

[0001] This application claims the benefit of and priority to U.S. Patent Application No. 17 / 033,512, filed September 25, 2020, which is incorporated herein by reference in its entirety for all purposes.

[0002] Plastic products are primarily disposable and often not recycled. Global annual production of plastic is approximately 350 million tonnes, of which about 10% is recycled, 12% is incinerated, and the remainder (78%) accumulates in landfills or the natural environment, taking approximately 500-1,000 years to decompose. Plastic production is expected to double by 2030 and triple by 2050.

[0003] Mechanical recycling is a dominant strategy for recycling plastics and involves crushing, melting, and re-extrusion of plastic waste. Because recycling facilities are often designed to process sorted material streams at high purity levels to preserve high levels of material performance in recycled products, high contamination rates and mixed material streams are a major cause of low yields and low value in the recycling process. Feedstock impurities reduce the effectiveness of recycling, even after several cycles of mechanical recycling, due to complex formulations containing additives and physical degradation of materials. For example, among plastic materials, polylactic acid (PLA) is a common waste plastic that often goes undetected in polyethylene terephthalate (PET) sorting and mechanical recycling operations. As another example, chlorinated compounds such as polyvinyl chloride (PVC) are not tolerated in both mechanical and chemical recycling operations because corrosive compounds are produced during the recycling process, limiting the value of hydrocarbon output. Summary of the Invention

[0004] Techniques (e.g., methods, systems, non-transitory computer-readable media storing code or instructions executable by one or more processors) are provided for managing the recycling of molecular components incorporated into waste materials.

[0005] In particular, techniques may be directed to characterizing the chemical or material identity of constituent materials entering waste streams. Constituent materials may be identified using chemical fingerprints derived from a comprehensive library including spectral information, physical properties, computational algorithms, and machine learning. Material characterization data may be used to develop chemical processes that convert materials into target products. Identification of target products and processes may be informed by logistics information, market data, and real-time exchange data. For example, a target product may be identified as one associated with relatively high market demand and that can be produced relatively efficiently using at least one constituent material.

[0006] In some embodiments, the method may include accessing characterization data of a feedstock. The characterization data may include one or more spectra collected according to one or more spectroscopy methods. The method may include using the characterization data to predict a set of constituent materials included in the feedstock. The method may include predicting a material composition of the feedstock using the predicted set of constituent materials. The method may include identifying one or more target products using the predicted material composition of the feedstock. The method may include generating a set of chemical reaction schemas that enable at least a partial conversion of the feedstock into one or more target products. The method may also include storing the identification of the material composition of the feedstock, the one or more target products, and the set of chemical reaction schemas in a data store.

[0007] In some embodiments, the method may also include identifying one or more inputs to a fitness function, the one or more inputs describing a chemical reaction scheme of the set of chemical reaction schemata. The method may also include generating an output of the fitness function using the one or more inputs. The method may also include selecting an implementation scheme from the set of chemical reaction schemata according to the fitness function, the one or more inputs, and the one or more target products. Identifying the one or more target products may include accessing inventory information describing the set of products and using the inventory information to identify an incomplete subset of the set of products as the one or more target products. The inventory information may include one or more of: a quantity of raw material available for conversion, or a quantity of the target product of the one or more target products available in the geographic region. The method may further include directing a portion of the raw material to a materials recycling facility configured to convert the portion of the raw material into at least one target product of the one or more target products. Generating the set of chemical reaction schemas may include accessing a chemical reaction inventory, the chemical reaction inventory including representations of chemical reactions describing the conversion of raw materials into one or more target products, and populating the set of chemical reaction schemas from the chemical reaction inventory. Generating the set of chemical reaction schemas may include simulating a first constituent reaction of a chemical reaction scheme of the set of chemical reaction schemas using a machine learning model. Generating the set of chemical reaction schemas may include estimating an output of a reward function, the output of the machine learning model serving as an input to the reward function. Generating the set of chemical reaction schemes may also include estimating a maximum value of the reward function by modifying an input to the machine learning model, the input being an output from a second constituent reaction that precedes the first constituent reaction in the chemical reaction scheme.

[0008] In some embodiments, a computer system includes one or more processors and memory in communication with the one or more processors, the memory configured to store computer-executable instructions, execution of which causes the one or more processors to perform one or more aspects of the methods described above.

[0009] In some embodiments, a computer-readable storage medium stores computer-executable instructions that, when executed, cause one or more processors of a computer system to perform one or more aspects of the methods described above. [Brief explanation of the drawings]

[0010] [Figure 1] 1 illustrates an exemplary technique for managing the recycling of molecular components of a feedstock, according to some embodiments of the present disclosure. [Figure 2] 1 illustrates an exemplary workflow for predicting the material composition of a feedstock, according to some embodiments of the present disclosure. [Figure 3] 1 illustrates an exemplary workflow for generating a set of chemical reaction schemas according to some embodiments of the present disclosure. [Figure 4] 1 illustrates an exemplary workflow for tuning a chemical reaction process using chemical and logistics data, according to some embodiments of the present disclosure. [Figure 5] 1 shows an example flow illustrating a method for managing the recycling of molecular components of a feedstock, according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Mechanical recycling has limited applicability to mixed, complex, and contaminated waste streams, in part due to the use of mechanical separation and reformation processes that may be insensitive to chemical contaminants and unable to alter the chemical structure of waste materials. Chemical recycling can address the limitations of mechanical recycling by breaking down the chemical bonds of waste materials into smaller molecules. For example, in the case of polymeric materials, chemical recycling can provide a means to recover oligomers, monomers, or even basic molecules from plastic waste feedstocks. In the case of polymers, chemical recycling processes can include operations to depolymerize and dissociate the chemical composition of complex plastic products so that their by-products can be upcycled into feedstocks for new materials.

[0012] Chemical recycling elements may allow materials to be repeatedly dissociated into their primary raw materials. In this way, rather than being limited to a limited number of physical processes by chemical structure and material integrity, as is the case with mechanical recycling, chemical recycling may be integrated into an "end-to-end" platform to facilitate the reuse of recyclable materials' molecular components. For example, chemical recycling products may include basic monomers (ethylene, acrylic acid, butyric acid, vinyl, etc.), feedstock gases (carbon monoxide, methane, ethane, etc.), or elemental materials (sulfur, carbon, etc.). Instead of being limited to a single group of recycled products, products that can be synthesized from intermediate chemicals that can be produced from the waste through chemical reactions may be identified based on the molecular structure of the input waste material. In doing so, the end-to-end platform may manage the waste stream by generating chemical reaction schemes that convert waste materials into one or more target products. For example, the end-to-end platform may direct waste feedstocks to a chemical recycling facility for chemical conversion of the waste material into the target product. In this way, the end-to-end platform may improve the implementation of reuse and recycling strategies and increase the diversion of waste materials from waste to the recycling system.

[0013] The end-to-end platform can collect data describing the waste material, molecular components, and quantities of finished products and use this information to actively manage the recycling process to produce the target product. Chemical reaction schemes can be modified or updated using the data to change target product quantities, endpoints, or chemical structures. For example, the conversion of waste to feedstock monomers and back to raw polymers can be tracked and integrated into local, regional, and / or global waste recycling or upcycling systems. The system can include recycling operators such as chemical processing equipment, material recycling facilities, waste sources, and purified polymer feedstock endpoints. In turn, waste sources can include, but are not limited to, industrial, institutional, or domestic waste sources. Downstream processing equipment can incorporate pure material products of chemical recycling without receiving repurposed waste materials.

[0014] Potential benefits of the chemical recycling processing schemes described herein include producing products from waste materials that are unavailable through mechanical recycling, which may increase the proportion of waste materials that can be recycled. For example, plastic feedstocks can be completely converted into non-plastic materials, such as hydrocarbon gases, which can then be synthesized into new, different polymeric materials. Furthermore, assisted chemical process development implementing machine learning capabilities can potentially reduce process development timelines and increase the efficiency of chemical recycling processes, making them viable on a large scale. For example, embodiments described herein may include accelerating the development timeline of new chemical reaction schemes from bench scale to pilot scale to ultimately industrial scale, which can typically take as long as 17 years, to real-time processes using multi-scale simulations of chemical recycling processes already active in logistics networks.

[0015] Further benefits may potentially address limitations of conventional recycling methods, which are typically designed to process relatively pure waste streams containing minimal contaminants. The techniques described herein may improve recycling processes by characterizing waste materials and managing recycling schemes to produce desired products with improved efficiency and performance. Benefits may include, but are not limited to, product yield per weight of waste material, energy consumption, environmental impact of the recycling process, or the percentage of recyclable waste diverted to landfills or disposed of in bodies of water.

[0016] 1 illustrates an exemplary workflow 100 for managing the reuse of molecular components of feedstocks, according to some embodiments of the present disclosure. Generally, workflow 100 may include one or more systems for characterizing waste materials, predicting the composition of waste materials, and developing chemical recycling protocols for waste materials, through which the waste materials may serve as feedstocks that are converted into target products by one or more chemical recycling processes.

[0017] In some embodiments, the workflow 100 may include a material characterization system 110, which may be implemented as a point-of-use device such as a tablet, smartphone, laptop computer, or dedicated sensor device that may include one or more sensor tools to facilitate spectroscopic, imaging, or chemical characterization of the waste material 111. The waste material 111 may be or include a material that can serve as a feedstock for a recycling process. For example, the waste material 111 may be or include a commonly recycled material such as polyethylene terephthalate (PET), such that the waste material 111 may be labeled before being characterized as a feedstock for a recycling process. In some cases, the waste material 111 may also include contaminants or additives that may be identified by analysis by the characterization system 110 and may inform the use of the waste material 111 as a feedstock, as described in more detail below. In some cases, the waste material 111 may be labeled with additional metadata to inform subsequent analysis of the material as part of the workflow 100. For example, the label may be or include a CAS number, which may allow standard characterization data to be searched or accessed from a database of standard data.

[0018] The material characterization system 110 may provide one or more types of characterization data 113 describing the waste material 111. The characterization data 113 may include spectroscopic data generated by measuring the interaction of one or more wavelengths of light with the waste material 111. For example, the characterization data 113 may include, but is not limited to, spectroscopy such as surface light reflectance / absorbance data 115, transmitted absorbance data 117, or hyperspectral image data measured by illuminating the waste material 111 with light within one or more spectral ranges. In some cases, the characterization data 113 may include infrared absorbance data, infrared reflectance data, visible absorbance or reflectance data, near-infrared data, ultraviolet absorbance data, or microwave or x-ray interaction data (e.g., x-ray fluorescence). In some embodiments, the characterization data 113 may include physical and chemical characterization, including, but not limited to, surface resistivity data, physical characterization data such as hardness or tensile properties, or other physical or chemical properties that may contain characteristic information for distinguishing the waste material 111 from other types of waste material.

[0019] In some embodiments, waste material 111 may include, but is not limited to, polymers, plastics, composites containing plastics, non-plastics, lignocellulosic materials, metals, glasses, and / or rare earth materials. Polymeric and plastic materials may include materials formed by one or more polymerization processes and may include highly cross-linked polymers as well as linear polymers. In some cases, waste material 111 may include additives or contaminants. For example, plastic materials may include plasticizers, flame retardants, impact modifiers, rheology modifiers, or other additives included in waste material 111, for example, to impart desired properties or enhance forming characteristics. In some cases, waste material 111 may incorporate constituent chemicals or elements that may not be compatible with a wide range of chemical recycling processes; therefore, characterization data 113 may include information specific to such chemicals. For example, decomposition of halogen- or sulfur-containing polymers may produce corrosive by-products, which may inhibit or impair chemical recycling of waste material 111 containing such elements. An example of a waste material 111 that contains halogen components is polyvinyl chloride (PVC). For example, the decomposition of PVC can produce chlorine-containing compounds that can act as corrosive by-products.

[0020] Once collected, the characterization data 113 may be accessed by a computer system 120 implementing one or more elements of the workflow 100. In some embodiments, the computer system 120 may include a server, one or more servers, a virtual machine, or multiple virtual machines, which may be implemented in a physical computer system or a distributed computer system (e.g., a cloud computing system). In some cases, the computer system 120 may communicate with one or more external systems, such as the material characterization system 110, over a network 130. The network may be a public network, such as the Internet, or a private network, such as a client network, a restricted network, or a local area network.

[0021] In some embodiments, computer system 120 may execute a process for predicting the set of constituent materials contained in waste material 111. As described in more detail below with reference to FIG. 2, computer system 120 may access a library of spectra corresponding to one or more spectroscopic methods. For example, database 131 may include spectroscopic data for multiple standards, material combinations, as well as empirical characterization data for real-world materials. In some embodiments, database 131 may communicate with computer system 120 via network 130. Additionally, computer system 120 may store at least a portion of the library of spectra in memory of computer system 120.

[0022] In some embodiments, predicting the set of components contained in the waste material 111 may include executing a material identification application 140. The material identification application 140 may include a spectral analyzer 143 that receives the characterization data 113. In some embodiments, the spectral data 141 may also serve as input to the spectral analyzer 143, which may be provided by accessing a library of spectra in the memory of the computer system 120 or from the database 130. In some embodiments, the spectral data 141 may be simulated or empirically measured. As described below, the material identification application 140 may identify one or more bands of interest in the characterization data 113 and use the one or more bands of interest as part of generating a chemical fingerprint 145 of the waste material 111. Generally, the chemical fingerprint 145 of the waste material 111 describes a set of characteristic information derived from the characterization data 113, which may identify material components of the waste material 111 that can, for example, be introduced as feedstock into a chemical recycling process.

[0023] In the context of the material identification application 140, the chemical fingerprint 145 may describe a prediction of the constituent materials and material composition that make up the waste material 111 based at least in part on the characterization data 113 and the spectral data 141. For example, the chemical fingerprint 145 may describe the major constituent compounds and additives or contaminants indicated by the characterization data 113. The chemical fingerprint 145 may also describe the relative composition of each constituent material that makes up the waste material 111, for example, if the characterization system 110 implements calibrated spectroscopy that facilitates absolute composition methods. In some embodiments, the material composition may be predicted based on standard data or as part of a machine learning model trained using a dataset that includes information from mixed materials, as described in more detail with reference to FIG. 2 below.

[0024] As described in more detail with reference to FIG. 3, chemical fingerprint 145 may enable identification of one or more target or desired chemical recycling products. For example, in some embodiments, computer system 120, as part of executing material identification application 140, may access chemistry inventory information describing one or more products, such as a set of products, that may be produced by a chemical recycling process that incorporates waste material 111 as a feedstock. For example, computer system 120 may identify a band of interest, from which computer system 120 may also provide one or more target or desired products that may be produced from waste material 111. Identification of target products may be facilitated by accessing chemical recycling process data, such as feedstock-product pairings, as in a searchable table (e.g., a lookup table), as described in more detail with reference to FIG. 3. By way of example, chemistry 145 may be used to predict that waste material 111 may be or may include a PET fingerprint with one or more additives or impurities that may eliminate one or more potential chemical recycling processes or result in balancing or adjusting feedstock ratios to enable recycling and / or reduced waste in the recycling process system. From this information, computer system 120 may access chemistry data to provide information describing one or more target products. In this example, computer system 120 may cross-reference the raw material chemistry data with the chemistry data of impurities, additives, and contaminants to reduce the likelihood of improperly identifying the target product.

[0025] Additionally or alternatively, computer system 120 may receive one or more desired product identifications from a user of computer system 120 as manual input based on chemical fingerprint 145. For example, the computer system may include a user interface or console application by which one or more users may interact with one or more applications of computer system 120. In some embodiments, the user interface may allow a user to review the data comprising chemical fingerprint 145, conduct a search for potential chemical recycling products, and indicate one or more desired products.

[0026] In some embodiments, computer system 120 may execute a chemical reaction modeling application 150, which may enable computer system 120 to simulate one or more chemical recycling processes in which waste material 111 acts as a feedstock for producing one or more target or desired products. Chemical reaction modeling application 150 may access one or more representations of chemical reactions describing the conversion of raw materials into target products, which may be stored in a database of chemical reaction data 151, as described in more detail below with reference to FIG. 3. The chemical reaction data may be or include a machine-searchable catalog of elementary chemical reactions for depolymerizing polymers, breaking covalent bonds in chemical reactants, or otherwise physically or chemically converting waste material 111 into target products.

[0027] In some embodiments, the database of chemical reaction data 151 can be or include a chemical reaction inventory, which can serve as an initial set of chemical reactions input into a chemical process simulation, as described in more detail with reference to FIG. 3 . Similar to database 131, the database of chemical reaction data 151 can be a network data store or memory device in the same physical location as computer system 120. In some cases, the chemical fingerprint 145 can serve as an additional input to the chemical reaction modeling application 150. For example, the chemical fingerprint 145 can include information describing the phase, structure, and amount of one or more raw materials and products, as described above. Thus, the input to the chemical reaction modeling application 150 can be or include input molecules, output molecules, catalysts, reagents, solvents, and chemical processing parameters, including, but not limited to, residence time, reaction temperature, reaction pressure, or mixing rate and pattern.

[0028] In some embodiments, chemical reaction modeling application 150 can be or include one or more unit operation models that can be implemented to simulate the constituent reactions of chemical reaction scheme 153. Chemical reaction modeling application 150 can generate multiple chemical reaction schemes 153, which can include different constituent reaction processes or describe different reaction products. In some embodiments, chemical reaction modeling application 150 can simulate one or more of the unit operation models using machine learning models, such as artificial neural networks implementing deep learning features, “black box” optimization techniques, supervised learning, reinforcement learning, or other standard machine learning approaches. In this manner, if chemical reaction scheme 153 includes multiple constituent reactions, as represented by a series of unit operation models, chemical reaction modeling application 150 can implement one or more machine learning models, where the output of a first model serves as the input of a second model. As described in more detail with reference to FIG. 3 , chemical reaction modeling application 150 can implement a model tuning protocol through a reward function, which can enable iterative modification of one or more parameters of a unit operation model to optimize or improve the model. In some embodiments, tuning the model may include estimating the output of a reward function as a function of one or more values ​​calculated by chemical reaction modeling application 150 and modifying one or more model parameters to maximize the output of the reward function. In addition to the reward function, training of the machine learning model implemented as part of chemical reaction modeling application 150 is described in more detail below.

[0029] In some embodiments, one or more unit operation models may be based on first principles rather than machine learning approaches. As an example, a chemical recycling process, such as a polymer catalyst cracking unit operation, may be simulated by chemical reaction rate equations, for which input variables may be supplied by, for example, a preceding unit operation model or by heuristics from a table lookup. In this manner, the set of unit operation models simulated by chemical reaction modeling application 150 may include both machine learning and first principles models. In some embodiments, if material characterization system 110 includes an online sensor system as part of the material classification process, chemical reaction modeling application 150 may progressively access or receive chemical fingerprint 145 data and update reaction schema 153 in response to receiving updated information. Real-time updates to chemical reaction simulations may improve the performance of the chemical recycling process managed by computer system 120. For example, waste material 111 may be redirected from an initial destination to another destination following an update to chemical reaction schema 153, which may improve one or more performance factors described below.

[0030] In some embodiments, the chemical reaction schemas 153 or constituent chemical unit operations may be filtered by one or more selection operations performed by the computer system 120. For example, as described in more detail below with reference to Figures 3-4, a fitness function may be defined by which an implementation scheme may be selected. The fitness function may be an object model with multiple inputs, which may include, but are not limited to, predicted input quantities, output quantities, energy input values, cooling water demand, material costs, or fuel consumption due to logistics operations involved in transporting the waste material 111. In some embodiments, the fitness function may receive derived values ​​as inputs, including, but not limited to, reaction yields, conversion efficiencies, chemical reaction selectivities, heat balance values, energy consumption, or environmental impacts. Environmental impacts may account for the generation of regulated by-products, including, but not limited to, greenhouse gases, chemical waste, or vitrification slag. For example, a "greenness" methodology may be used to establish integrated metrics that enable a comprehensive quantitative measurement of the environmental impact and sustainability of proposed reaction conditions. Similarly, "green chemistry principles and life cycle assessments" may be used to promote safe processes that minimize the generation of hazardous substances. In some embodiments, each parameter provided to the fitness function may be given a weight that may influence the desirability of a given chemical reaction scheme 153 or constituent chemical unit operation.

[0031] The chemical reaction modeling application 150 may provide output, including the chemical reaction schema 153, to the optimization engine 160. The optimization engine 160 may be or may include a machine learning model and may facilitate real-time modification or selection of the chemical reaction schema 153 based on inputs, including, but not limited to, those generated by the chemical reaction modeling application 150, the chemical fingerprint 145, or inventory information 163. In some embodiments, the inventory information 163 may be accessed from a networked system of recycling information 161. The recycling information 161 may be stored in a database that is updated progressively, such as in real time, and the database may detail the material supply chain and track waste materials through their decomposition and subsequent recombination into new materials. For example, the inventory information 163 may include the quantity or quality of materials available in a logistics network that may correspond to a geographic region. Similarly, the inventory information 163 may include inventory information for target products available within a geographic region.

[0032] In some embodiments, the optimization engine 160 may use inventory information 163 to modify the target or desired products that serve as input to the chemical reaction modeling application 150. For example, the computer system 120 may access the inventory information 163. Using the inventory information 163, the computer system 120 may identify a larger subset of target products to limit the number of chemical reaction schemas 153 generated. As an example, waste material 111 may be identified as a potential feedstock for several chemical recycling methods that offer multiple possible reaction products. Through accessing the inventory information 163 corresponding to the possible reaction products, a selection of one or more of the possible reaction products may be made for a target product with limited supply, avoiding oversupply of an already popular product or generating a product that reflects the potential for high consumption demand. In some embodiments, as described in more detail below with reference to FIG. 4, a real-time exchange may connect recyclers, chemical companies, and other consumers or producers of recyclable materials. The real-time exchange may implement inventory planning, supply and demand management, recycled product markets, and logistics management. For example, computer system 120 may direct waste material 111 or a portion of waste material 111 to a materials recycling facility or other processing operation where waste material 111 may be converted into one or more target products. An example of directing waste material 111 may include identifying a recipient facility and a sender facility and generating logistics information that may be provided to the recipient facility or sender facility.

[0033] As part of implementing a chemical recycling process for waste material 111, computer system 120 may provide one or more of chemical reaction schema 153, model output, chemical fingerprint 145, characterization data 113, or other information generated, processed, or accessed by computer system 120 to external computer system 170. External computer system 170 may be or include a control server at the materials recycling facility. For example, computer system 120 may receive characterization data 113 from an on-site characterization system 110 including multiple sensors and probes, generate chemical reaction schema 153 as described above, and provide chemical reaction schema 153 and / or implementation schema to external computer system 170 for execution using a chemical process unit at the materials recycling facility. In this manner, external computer system 170 may receive information from computer system 120 via network 130.

[0034] In some embodiments, computer system 120 may store the same or similar information in a data store, such as database 131 or chemical reaction inventory database 151. For example, computer system 120 may manage a chemical reaction inventory by storing a representation of all chemical inputs and outputs of reactions, as well as the catalysts and reaction conditions involved. The information stored in the chemical reaction inventory may then be used to optimize known, widely used reactions and to aid in the exploration and discovery of novel catalysts and reaction conditions that may be applicable to decomposing plastic waste. Similarly, bands of interest developed by material identification application 140 may be stored for training and improving material fingerprinting approaches, as described in more detail below with reference to FIG. 2.

[0035] FIG. 2 illustrates an exemplary workflow 200 for predicting the material composition of a feedstock, according to some embodiments of the present disclosure. As part of developing and managing a chemical recycling process for waste materials that serve as feedstocks (e.g., waste material 111 in FIG. 1 ), workflow 200 may incorporate a variety of data sets, data processing techniques, and analytical operations. A system implementing workflow 200 may be or include a computer system (e.g., computer system 120 in FIG. 1 ) described with reference to FIG. 1 , or may be a communications system, such as a client device hosting a spectral analysis application (e.g., material identification application 140 in FIG. 1 ). Output data from workflow 200 may facilitate the prediction of chemical reaction schemas, as described below, for example, by generating chemical fingerprint data that can be used to identify target products, select candidate chemical reaction unit operations, or collect inventory information from a logistics network, among other uses described above with reference to FIG. 1 .

[0036] The operations of workflow 200 may include one or more data capture and processing operations. In some embodiments, a spectral database 210, which may be an example of database 131 in FIG. 1 , may store and process spectral data 220 and associated data 230 as part of the identification of experimental materials and compositions by a spectral analysis 240 system. The spectral data 220 may be or include calibrated or uncalibrated spectroscopic data prepared to facilitate the generation of chemical fingerprint data 250 by the spectral analysis system 240. The spectral data 220 may include spectroscopic characterization data of a pure control 221. The pure control may include a base polymer film without additives or contaminants, also referred to as a spectroscopic standard, which may be, or for several individual films that may include, polypropylene, polyethylene, polystyrene, high-density polyurethane, low-density polyurethane, polyethylene terephthalate, acrylonitrile butadiene styrene, polycarbonate, or polyamide. Additionally, the spectroscopic characterization data of the pure control 221 may include, but is not limited to, spectroscopic data of a control polymer film with a known amount of additive, or even the additive alone. A diverse set of spectral data can also be generated for the control sample using multiple modalities across the electromagnetic energy spectrum (X-ray fluorescence, radio frequency, near infrared, short wave infrared, mid wave infrared, THz, or mm range), as described above. The molecular and elemental components of the control material can also be characterized and included in the spectroscopic characterization data of the pure control 221.

[0037] In some embodiments, the spectral data 220 can be or can include spectroscopic characteristics of a material sample 223, which can include data generated by one or more spectroscopic techniques applied to samples of waste materials or their components. For example, decomposition techniques can be applied to further estimate ground truth information about the waste material by using destructive methods such as gas chromatography mass spectrometry (GCMS), laser-induced breakdown spectroscopy (LIBS), or non-destructive methods such as attenuated total resonance (ATR)-fast infrared (FTIR). The spectral data 220 can include spectroscopic characterization of a material sample 221 collected from a recycling network and progressively provided to the spectral database 210. For example, a network of material recycling facilities can collect material characterization data (e.g., characterization data 113 of FIG. 1 ) as part of the waste material intake or sorting process and provide the data to the spectral database 210 as part of implementing management of chemical recycling across the network.

[0038] The spectral data 220 may be coordinated with the related data 230, for example, through labeling the spectral data with nomenclature data 231, formula data 233, or other metadata 235. In some cases, the related data 230 may facilitate spectral analysis techniques implemented by the spectral analysis system 240 in response to the spectral data 220, including, but not limited to, model training techniques described below. The spectral data 220 and the related data 230 may be stored in separate data stores connected to the spectral database 210 via a network. For example, the spectral data 220 may be accessed by the spectral database 210 as part of a distributed data system operation, such as an extract, transform, and load (ETL) process. Similarly, the related data 230 may be collected in one or more databases located at one or more physical locations and accessed or received by the spectral database 210.

[0039] Data acquisition 211 operations may be applied to the spectral data 220 and associated data 230 in preparation for subsequent data processing. For example, data acquisition 211 may include one or more data transformations, such as ETL processes, that may modify the format or representation of the data. For example, a spectral file 213 from the spectral data 220 may be combined with associated associated data 215 as part of data acquisition 211 to generate spectral data entries in a standardized format 217. This may include converting the spectral file 213 from a standard data format, such as a comma-separated value format, to a key-value pair format. The key may be or include a searchable database label, such as a unique identifier. The standardized format 217 may include fields for associated data 215, such as labels related to the chemical composition of the sample. Examples of labels may include, but are not limited to, name data 231, formula data 233, molecular weight data, and associated metadata 235, such as SMILES string data, MOL file data, CAS number, or structure representation.

[0040] As shown in FIG. 2 , the spectral analysis system 240 can access data in the standardized format 217 as part of generating the chemical fingerprint data 250. The spectral analysis system 240 can generate the chemical fingerprint data 250 for the standard control material and the characterized waste material and can incorporate both automated and manual analysis techniques. For example, the spectral analysis system 240 can implement a set of visualization tools 241 and implement machine learning methods or other computational spectral analysis 245 techniques as part of developing the chemical fingerprint data 250. The visualization tools 241 can be used to query relevant spectra for specific materials, additives, or contaminant chemicals. In some embodiments, the visualization tools 241 can display and align sample spectra with one or more control spectra for comparison by human eye or by machine image analysis (e.g., by a convolutional neural network trained to classify the spectra). The visualization tools 241 can enable rapid analysis of anomalous spectra as well as manual curation of spectral datasets used as training sets for machine learning. The visualization tools 241 may allow for curated labeling of sample spectra, such as when used to refine training sets in reinforced learning.

[0041] In some embodiments, the spectral dataset may be normalized through data preprocessing 243, including a modular normalization approach. For example, intensity normalization may be applied to the raw spectral data as part of identifying a band of interest 251, e.g., based on identifying key features such as peaks or bands. As described below, the band of interest 251 may serve to guide a material characterization system by configuring a spectroscopic probe and may additionally or alternatively be used in identifying unlabeled spectra detected by broadband characterization techniques. For example, the band of interest 251 for a transmittance scanner may be or include 1620-1787 nm from 1350-2450 nm, such that normalization or other data processing may be preferentially applied thereto. Similarly, the band of interest 251 for a reflectance scanner may be or include 1117-1261 nm from 900-1700 nm. Normalization may refer to intensity normalization and may be applied in situations including, but not limited to, when the spectroscopic sensor device is not intensity calibrated.

[0042] Data processing 243 may include baseline and other compensation techniques. For example, a baseline in the spectral file may be detected, which may correspond to background signals or general trends in the raw spectra that are not attributable to the measured sample. In some cases, normalization may involve multiple operations, including, but not limited to, baseline subtraction and intensity normalization, which is performed by dividing the intensity data obtained in each band by the sum of all differences. In this way, the processed spectral data may be normalized across different material thicknesses and transparencies. The normalized spectrum may be smoothed to improve subsequent computational spectral analysis 245.

[0043] In some embodiments, the processed spectral data may serve as training data 247 for a machine learning model implemented as part of computational spectral analysis 245. For example, the machine learning model may be or include a support vector machine (SVM) classifier. The machine learning model may be trained by model training 249 techniques, including but not limited to supervised learning or reinforcement learning, using at least a portion of the normalized spectral data, which may be labeled or unlabeled. In some embodiments, model training 249 may be implemented using sub-band data, which may provide improved classification accuracy relative to training using the full spectrum. Model training 249 may implement an adversarial learning approach, such as a discriminator, which may train the machine learning model implemented as part of computational spectral analysis 245.

[0044] The normalized spectral data may be processed by trained machine learning models or other computational methods, such as procedural or rule-based models, to look for patterns in the signal related to material signatures 253, additive or contaminant signatures 255, or other information indicative of chemical type, composition, morphology, structure, or purity. For materials incorporating multiple different additives, contaminants, or impurities into the primary material, such as units of different forms of recycled PET objects containing various plasticizers, such as those received by a materials recycling facility, multiple regions covering the peak signal of the material may be identified as bands of interest 251. In some embodiments, as many as 30-40 bands of interest 251 may be selected, excluding less informative bands that may be common across all forms of recycled raw materials. In an example, a classifier implementing an SVM trained to classify materials may be provided with bands of interest 251 for a waste material sample based on spectral labels included in data acquisition 211.

[0045] In some embodiments, chemical fingerprint data 250 may be stored in fingerprint database 260. The fingerprint database may be in communication with spectral analysis system 240, for example, over a network or in the same physical location as spectral analysis system 240. As part of performing computational spectral analysis 245, spectral analysis system 240 may access chemical fingerprint data 250 stored in fingerprint database 260. For example, as described below with reference to FIG. 3 , an SVM trained by model training 249 may use access to bands of interest 251 and material signatures 253 for a broad class of materials, such as polymers, to provide classification with sufficient accuracy to distinguish between different polymer structures, side chains, backbones, or other information that may affect the identification of potential target products, as well as the chemical reaction formulations that transform the materials whose spectral data have been classified into potential target products.

[0046] FIG. 3 shows an example workflow 300 for generating a set of chemical reaction schemas according to some embodiments of the present disclosure. As part of managing a chemical recycling process, a computer system (e.g., computer system 120 of FIG. 1 ) may simulate one or more chemical recycling unit operations as part of unit operation simulation 310. Unit operation simulation 310 may receive data generated by a material identification and characterization application, as described above, as part of a platform for guiding the chemical recycling process. Workflow 300 may include the implementation of machine learning and rule-based models as part of generating a set of reaction conditions describing a chemical process for converting a feedstock into a target product. The feedstock may be waste material received by a materials recycling facility. The target product may be identified by the computer system as part of workflow 300 or may additionally or alternatively be specified by external input.

[0047] The unit operation simulation 310 may receive or access material identification data 320 (e.g., chemical fingerprint data 250 of FIG. 2 ) for the waste material, so that the waste material's identity and composition may serve as input 321. The material identification data 320 may include desired inputs 323, which may be provided by an external system, such as via an exchange system described below with reference to FIG. 4 . The unit operation simulation 310 may also receive input from a chemical reaction inventory 330. The chemical reaction inventory 330 may store representations, such as embeddings, of all chemical inputs 333 and outputs 335, catalysts 337, and reaction conditions 331 of the reactions involved. Reaction models 339 may also be stored as part of the chemical reaction inventory 330, which may enable the unit operation simulation 310 to include rule-based reaction models in addition to machine learning approaches as part of the guided chemical recycle 350 formulation. The inputs 333 and outputs 335 can be cross-referenced in a chemical reaction inventory 330 so that material identification data 320 can be paired with potential target products, which can serve to define an initial set of chemical reaction schemata that can be optimized, as described below.

[0048] As part of generating optimized reaction conditions 340, unit operation simulation 310 may modify known, widely used reactions and aid in the exploration and discovery of novel catalysts, reagents, or solvents 343 and reaction conditions 341 that may be applicable to decompose waste materials. In some embodiments, unit operation simulation 310 may incorporate molecular modeling techniques, such as density functional theory and molecular dynamics, with a known set of catalysts or reagents to formulate new catalyst data 337 not previously present in chemical reaction inventory 330. Unsupervised ML algorithms, including various clustering methods, Gaussian mixture models, factor analysis, and learning reaction embedding through deep neural networks (DNNs), can be applied to data from chemical reaction inventory 330. In some embodiments, supervised ML algorithms, such as regression models or DNNs, can be used to improve chemical reaction models 339. As described in the context of spectral analysis (e.g., spectral analysis system 240 of FIG. 2), machine learning approaches implemented as part of unit operation simulation 310 may be trained using datasets from chemical reaction inventory 330, which may include one or more preprocessing steps such as labeling, curation, or other approaches to select training data and guide the development of ML models.

[0049] In searching for recommended catalysts and chemical reactions, multiple approaches may be incorporated into the guided chemical recycle 350 simulation to generate optimized reaction conditions 340. In some embodiments, one or more chemical processes may be simulated as a series of reaction models 353a-n, each receiving inputs 351a-n and producing outputs 355a-n. Each reaction model 353 may represent a chemical unit operation that forms a stage in the chemical recycle process. In some cases, a final reaction model 335n may output a final output that may represent a target product, as identified from the chemical reaction inventory 330.

[0050] In some embodiments, simulation results representing intermediate reaction conditions may be provided to an online learning algorithm to fine-tune the model and simulation technique. The online learning algorithm may incorporate a reward function 360 that may indicate the success of a reaction or reaction scheme. In some embodiments, the reward function 360 may generate threshold criteria representing one or more chemical process parameters, such as inputs 351 or outputs 355, by which the optimality of the overall induced chemical recycle simulation 350 may be determined. For example, one or more inputs 351 or outputs 355 may be provided to the reward function 360 at each iteration, and the unit operation simulation 310 may increment those inputs 351 or outputs 355 until a desired result is achieved, such as the output of the reward function 360 crossing a threshold, which may indicate that the reaction conditions from the induced chemical recycle 350 are optimized. In some embodiments, the output of the reward function 360 crosses a threshold, which may indicate that the reaction conditions from the induced chemical recycle 350 are above a minimum acceptable level rather than at a maximum level.

[0051] In some cases, the result may be or include an optimized chemical reaction scheme, such as a pyrolysis process, for efficiently breaking down plastic into a desired series of molecules. In another embodiment, reward function 360 may receive pressure, temperature, and catalyst levels for a Fischer-Tropsch reaction, given inputs of carbon monoxide and hydrogen, to produce liquid hydrocarbons that can be used as feedstock for plastics. Reward function 360 may be or include a general optimization algorithm, such as steepest descent, to derive increments for inputs 351 and outputs 355. Algorithms specific to each type of chemical reaction problem may also be included, depending on the desired input / output data or conditions requiring optimization. For example, simulation of a thermochemical process, such as pyrolysis, where sufficient process data is available, may use a different optimization technique, such as reinforcement learning. Reward function 360 may form part of the ML framework of workflow 300, such as reinforcement learning or black / gray-box optimization techniques, which may be used to guide the learning process and evaluate the learning results.

[0052] Reward function 360 may receive numerous inputs other than those included as inputs 351 to reaction model 353. For example, derived values ​​such as yield, selectivity, feasibility, energy usage, or environmental impact may serve as inputs to reward function 360. As an example, yield may be used to determine how much of a plastic polymer can be successfully converted into its constituent monomers and how much can be converted into unusable by-products. Similarly, selectivity may represent the ratio of desired monomer products to undesired reaction products. Feasibility may be used to capture the concept of whether proposed reaction conditions are feasible / practical to set up or run, demonstrating that some inputs to reward function 360 may be qualitative assessments based on non-physical criteria. Weights can be assigned to inputs to the parameters of reward function 360 to bias the reward function (and the learning process) toward a particular goal or objective. For example, if there is a constraint on selectivity for a particular reaction type, a higher weight can be assigned to selectivity in the calculation of the threshold criteria. In other cases, yield may be more important and assigned a higher weight.

[0053] The workflow 300 may include multiple outputs in addition to the optimized reaction conditions, which may improve the implementation, adoption, and performance of the chemical recycling process. For example, the optimized reaction conditions 340 may be visualized as a Markov process simulation 370, whereby the various stages of the chemical reaction scheme 345 are visualized as steps in a Markov process.

[0054] A Markov process, in general terms, represents different stages in a logistics or process chain as nodes connected by directional arrows, with a visual or quantitative indication of the weight of the connections between each node. In this manner, the Markov process simulation 370 may generate and / or present a dynamic visualization of the reaction scheme 345 to demonstrate the holistic effect of fine-tuning the constituent reactions of the reaction scheme 345 on the entire recycling pipeline. Thus, the Markov process simulation 370 may receive as input logistics data describing a materials recycling supply chain, as described in more detail below with reference to FIG. 4.

[0055] FIG. 4 illustrates an exemplary workflow 400 for adjusting a chemical reaction process using chemical and logistics data, according to some embodiments of the present disclosure. Elements of workflow 400 may include data provided by the logistics networks and databases described in previous figures, both of which may serve as inputs to optimization engine 410. Optimization engine 410 may implement one or more approaches to modify or adjust the chemical recycling process simulated by the systems of the preceding figures to better align with network factors, such as material inventory levels, logistics constraints, or consumption trends, as described below. The output of workflow 400 may be returned as input to one or more of the preceding systems, for example, during an iteration of unit operation simulation 310 of FIG. 3 .

[0056] In some embodiments, optimization engine 410 may receive data that can be broadly categorized as chemical recycling process data, such as material identification data 420 (e.g., chemical fingerprint data 250 in FIG. 2 ) or optimized reaction conditions 430 (e.g., optimized reaction conditions 340 in FIG. 3 ), and logistics data, such as material inventory 440 data, utilization data 450, or real-time data 460. Optimization engine 410 may apply one or more computational approaches to modify aspects of optimized reaction conditions 430 and output optimized data 470. In some embodiments, optimization engine 410 may receive additional input provided by exchange software 480, which may provide a platform for interaction with a network of entities 490 that produce raw materials, consume products, or produce chemical recycling intermediate materials, including, but not limited to, catalysts, solvents, or other consumables.

[0057] Materials inventory data 440 may include, but is not limited to, inventory data describing molecules 441, which may describe products or by-products, raw materials 443, chemicals 445, such as consumables, catalysts, or other reactants, or general materials 447. General materials may include, but are not limited to, other materials that may be used in operating the chemical recycling process (e.g., electricity, cooling water, heating fuel, or compressed gas). In this manner, materials inventory 440 may represent one or more constraints on the operation of the chemical recycling process identified in optimized chemical reaction conditions 430. Thus, information reflected in materials inventory 440 may potentially adversely affect optimized reaction conditions, for example, when a supply of a rate-limiting catalyst is unavailable.

[0058] Similarly, utilization data 450 may reflect local or regional trends in the chemical recycling infrastructure, which may promote or demote a reaction scheme (e.g., reaction scheme 345 of FIG. 3 ). For example, utilization data 450 may include, but is not limited to, data reflecting downstream demand 451, upstream supply 453, market data 453, or logistics data 455. Such supply, demand, and market factors may enable the optimization engine to adjust one or more parameters of the chemical recycling process simulation to reflect economic factors in addition to the physical and chemical factors reflected by the reaction model and thermochemical optimization. For example, a reaction scheme may produce a product whose supply already exceeds demand and for which there is a shortage of warehouse capacity in the logistics network. In such cases, the optimization engine 410 may demote the reaction scheme or identify a subset of target products 471. The subset of target products 471 may then be fed back into the chemical process simulation (e.g., workflow 300 of FIG. 3 ) to adjust the optimized reaction conditions 430.

[0059] The optimized reaction conditions 430 may describe specific conditions for a single chemical reaction scheme, as described with reference to Figure 3. However, the optimization engine 410 may simultaneously receive or access multiple chemical reaction schemes as part of selecting an implementation scheme or multiple implementation schemes, as described with reference to Figure 1. For example, identifying a subset of the target product 471 may enable selection of an implementation scheme that produces the subset of the target product 471.

[0060] Similarly, optimization engine 410 may output optimized logistics data, which may describe sources of raw materials 443 and recipients of products produced by optimized reaction conditions 430. For example, logistics data may include real-time data 460, which may include, but is not limited to, data describing materials recycling facility (MRF) 461 operation, distributed collection 463 operation, supply chain 465 status, or material characterization sensor data 467, the last of which may describe materials arriving at the MRF process in real time. In contrast, distributed collection 463 data may describe diverse sources of waste raw materials, such as industrial, commercial, institutional, and residential sources. Real-time data 460, combined with input from exchange software 480, may enable optimization engine 410 to specify product recipients and sources of raw materials for entities participating in a chemical recycling network (e.g., through network of entities 490).

[0061] The optimization engine 410 may implement a fitness function, including one or more computational techniques, such as a rule-based model or a machine learning model, to input multiple types of chemical recycling process data and available logistics data and generate a fitness value for the optimized reaction conditions 430. Similar to the reward function described with reference to FIG. 3 (reward function 360 in FIG. 3), the optimization engine 410 may receive weighted inputs, the weights of which may be externally specified by a human operator or an autonomous system, or may be developed by training the optimization engine 410 if a machine learning approach is employed. For example, the optimization engine may include an artificial neural network trained on a set of training data, which may be developed from historical operating data collected for a given chemical recycling process. The training allows the optimization engine to develop weights for inputs corresponding to process sensitivity to various possible logistics data, such as material inventory 440 or market data 455. For example, inventory data for a rate-limiting catalyst may significantly affect the feasibility of a chemical reaction scheme. In such a case, the weight for an input describing the supply of catalyst may be higher than the weight for an input with less impact. In the context of a loss function, the optimization engine may operate by minimizing the value of a loss function that is defined as the output of a machine learning model that receives chemical recycling process data and logistics data.

[0062] FIG. 5 shows an example flow diagram illustrating a method 500 for managing the recycling of molecular components of a feedstock, according to some embodiments of the present disclosure. As described with reference to FIGS. 1-4 , one or more operations constituting the method 500 may be performed by a computer system (e.g., computer system 120 of FIG. 1 ) in communication with additional systems, including, but not limited to, a characterization system, a network infrastructure, a database, and a user interface device. In some embodiments, the method 500 includes an operation 510 in which the computer system accesses characterization data for the feedstock. The characterization data (e.g., characterization data 113 of FIG. 1 ) may be generated by in-situ spectroscopic techniques, such as reflectance spectroscopy, transmission spectroscopy, or fluorescence spectroscopy, in one or more wavelength modalities, as described above. Furthermore, the characterization data may include physical or chemical information based on one or more different techniques, examples of which include hardness, tensile properties, or thermal phase properties. The characterization data may be provided to the computer system via a network (e.g., network 130 of FIG. 1 ).

[0063] In some embodiments, method 500 includes an operation 520 in which a computer system predicts a set of constituent materials contained in the feedstock. The computer system may implement a spectral analysis approach, as described in more detail with reference to FIG. 2, to identify chemical fingerprint data (e.g., chemical fingerprint data 250 of FIG. 2). This may include receiving standard and control data from a database of spectra (e.g., database 131 of FIG. 1). Additionally, operation 520 may include one or more implementations of a data transformation operation (e.g., data acquisition 211 of FIG. 2), a machine learning model, and the machine learning model may be trained using preprocessed training data (e.g., training data 247 of FIG. 2) prepared using spectral analysis techniques, including, but not limited to, normalization, baseline subtraction, or smoothing.

[0064] In some embodiments, method 500 includes an operation 530 in which a computer system predicts the material composition of the feedstock. The material composition of the feedstock may include information about the relative predominance of certain constituent materials in the feedstock, as opposed to the constituent materials. For example, the spectral analysis described with reference to FIG. 2 may identify certain constituent materials based on spectral features such as material signatures (e.g., material signature 253 in FIG. 2) or additive signatures (e.g., additive signature 255 in FIG. 2). However, such spectral analysis may not be able to distinguish between major components and impurities, for example, when the sensor is not intensity calibrated. Thus, in some cases, cross-referencing the spectral fingerprint with control data or training a machine learning model with the composition data may provide a predicted composition, such as the gravimetric composition of the feedstock.

[0065] In some embodiments, method 500 includes an operation 540 in which the computer system identifies one or more target products. Identification of the target products may be facilitated by a chemical reaction inventory (e.g., chemical reaction inventory 330 of FIG. 3 ), which may enable the computer system to identify a set of candidate products for the feedstock. For example, a material composition may include information about a primary component, which may be a polymeric material, and the chemical reaction inventory may describe multiple outputs (e.g., outputs 335 of FIG. 3 ) that can be produced by chemical recycling of the feedstock. Similarly, information describing a catalyst (e.g., catalyst 337 of FIG. 3 ) may describe contaminants that may contaminate the catalyst and thus eliminate each chemical reaction and its products from the set of candidate products. As described above, halogen-containing plastics (e.g., chlorine- and fluorine-containing plastics) may produce corrosive by-products that may exclude them from certain types of chemical recycling. As described above, the set of candidate products may be refined in one or more ways using logistics data (e.g., material inventory 440 of FIG. 4 ), which may enable identification of a limited number or an incomplete subset of target products.

[0066] In some embodiments, method 500 includes an operation 550 in which a computer system generates a set of chemical reaction schemata. Based on the material composition and the target product, the computer system may generate tailored reaction schemata using techniques described above with reference to FIG. 3 (e.g., workflow 300 of FIG. 3). For example, a chemical recycling process may be simulated as a series of unit operations represented as reaction models (e.g., reaction models 353a-n of FIG. 3). In some cases, a reaction model may receive as input the output of a preceding reaction model in the series, such as when a unit operation forms a stage in a process flow. As described in more detail with reference to FIG. 3, a unit operation simulation (e.g., unit operation simulation 310 of FIG. 3) may be tuned by a reward function (e.g., reward function 360 of FIG. 3), which may allow multiple factors to affect the operation of a given reaction scheme during tuning. For example, the reward function may receive as input chemical and physical information such as cooling water source capacity, fuel consumption information, environmental impact parameters, or other inputs that may directly affect the operation of the constituent process unit reaction models. Additionally, the reward function may allow the unit operation simulation to optimize for derived values ​​including, but not limited to, yield, selectivity, or efficiency.

[0067] In some embodiments, method 500 includes operation 560, in which the computer system stores the identification of the feedstock material composition, one or more target products, and the set of chemical reaction schemas. Output generated by the computer system may include, but is not limited to, reaction schemas, visualization information (e.g., Markov process simulation 370 of FIG. 3 ), as well as material compositions, materials of construction, and other predicted and generated data. In some embodiments, such generated data may be stored by the computer system in a data store, transmitted to an external computer system (e.g., external computer system 170 of FIG. 1 ), or returned as feedback data during simulation iterations. Additionally, material identification data, reaction schema data, target product data, or other generated information may be stored for later use in model training in one or more stages of method 500.

[0068] In the preceding description, various embodiments have been described. For purposes of explanation, specific configurations and details have been set forth to provide a thorough understanding of the embodiments. However, it will be apparent to those skilled in the art that embodiments may be practiced without the specific details. Additionally, well-known features may be omitted or simplified so as not to obscure the described embodiments. While the exemplary embodiments described herein focus on polymeric materials, these are meant to be non-limiting exemplary embodiments. Embodiments of the present disclosure are not limited to such materials; rather, they are intended to address material processing operations in which a wide range of materials serve as potential feedstocks for material recycling and / or upcycling processes. Such materials may include, but are not limited to, metals, biopolymers such as lignocellulosic materials, viscoelastic materials, minerals such as rare earth-containing materials, and complex composite materials or devices.

[0069] Some embodiments of the present disclosure include a system including one or more data processors. In some embodiments, the system includes a non-transitory computer-readable storage medium including instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes and workflows disclosed herein. Some embodiments of the present disclosure include a computer program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein.

[0070] The terms and expressions which have been employed are used as terms of description and not of limitation, and there is no intention in the use of such terms and expressions to exclude any equivalents of the features shown and described, or portions thereof, but it is recognized that various modifications are possible within the scope of the invention as claimed. Thus, although the invention as claimed has been specifically disclosed by embodiments and optional features, it will be understood that modifications and variations of the concepts disclosed herein may be effected by those skilled in the art, and that such modifications and variations are deemed to be within the scope of the invention as defined by the appended claims.

[0071] The description provides only preferred exemplary embodiments and is not intended to limit the scope, applicability, or configuration of the present disclosure. Rather, the following description of preferred exemplary embodiments will provide those skilled in the art with an effective description for implementing various embodiments. It will be understood that various changes can be made in the function and arrangement of elements without departing from the spirit and scope of the appended claims.

[0072] Specific details are given in the description to provide a thorough understanding of the embodiments. However, it will be understood that the embodiments may be practiced without these specific details. For example, certain computational models, systems, networks, processes, and other components may be shown as components in block diagram form in order to avoid obscuring the embodiments in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the embodiments.

Claims

1. 1. A method comprising: accessing characterization data for the raw material, the characterization data including one or more spectra collected according to one or more spectroscopic methods; using the characterization data to predict a set of constituent materials contained in the feedstock; predicting a material composition of the feedstock using the predicted set of constituent materials; identifying one or more target products using, at least in part, the predicted material composition of the raw materials; generating a set of chemical reaction schemas that enable the conversion of at least a portion of the feedstock into the one or more target products; storing in a data store an identification of the material composition of the feedstock, the one or more target products, and the set of chemical reaction schemas.

2. identifying one or more inputs to a fitness function, the one or more inputs describing a chemical reaction schema of the set of chemical reaction schemas; using the one or more inputs to generate an output of the fitness function; The method of claim 1 , further comprising: selecting an implementation scheme from the set of chemical reaction schemata according to the fitness function, the one or more inputs, and the one or more target products.

3. identifying the one or more target products accessing inventory information describing a set of products; and using the inventory information to identify an incomplete subset of the set of products as the one or more target products.

4. 4. The method of claim 1, further comprising directing a portion of the feedstock to a materials recycling facility configured to convert the portion of the feedstock into at least one target product of the one or more target products.

5. predicting the set of constituent materials contained in the raw material, accessing a library of spectra and associated metadata corresponding to the one or more spectroscopic techniques; identifying bands within the one or more spectra of the characterization data; and matching said bands to spectra in said library of spectra to predict constituent materials of said set of constituent materials.

6. generating the set of chemical reaction schemas, accessing a chemical reaction inventory, the chemical reaction inventory including representations of chemical reactions describing the transformation of the raw materials into target products of the one or more target products; and populating the set of chemical reaction schemas from the chemical reaction inventory.

7. generating the set of chemical reaction schemas, simulating a first constituent reaction of a chemical reaction scheme of the set of chemical reaction schemes using a machine learning model; estimating an output of a reward function, wherein an output of the machine learning model serves as an input to the reward function; 7. The method of claim 1, comprising: estimating a maximum value of the reward function by modifying an input to the machine learning model, wherein the input is an output from a second constituent reaction that precedes the first constituent reaction in the chemical reaction scheme.

8. 1. A system comprising: a memory configured to store computer-executable instructions; one or more processors in communication with the memory and capable of executing the computer-executable instructions, accessing characterization data for the raw material, the characterization data including one or more spectra collected according to one or more spectroscopic methods; using the characterization data to predict a set of constituent materials contained in the feedstock; predicting a material composition of the feedstock using the predicted set of constituent materials; identifying one or more target products using, at least in part, the predicted material composition of the raw materials; generating a set of chemical reaction schemas that enable the conversion of at least a portion of said feedstock into said one or more target products; and storing an identification of the material composition of the feedstock, the one or more target products, and the set of chemical reaction schemas in a data store.

9. The computer-executable instructions cause the one or more processors to: identifying one or more inputs to a fitness function, the one or more inputs describing a chemical reaction schema of the set of chemical reaction schemas; using the one or more inputs to generate an output of the fitness function; 9. The system of claim 8, further comprising: selecting an implementation scheme from the set of chemical reaction schemata according to the fitness function, the one or more inputs, and the one or more target products.

10. identifying the one or more target products accessing inventory information describing a set of products; and using the inventory information to identify an incomplete subset of the set of products as the one or more target products.

11. 11. The system of any one of claims 8-10, wherein executing the computer-executable instructions further causes the one or more processors to direct a portion of the feedstock to a materials recycling facility configured to transform the portion of the feedstock into at least one target product of the one or more target products.

12. predicting the set of constituent materials contained in the raw material, accessing a library of spectra and associated metadata corresponding to the one or more spectroscopic techniques; identifying bands within the one or more spectra of the characterization data; and matching said bands to spectra of said library of spectra to predict constituent materials of said set of constituent materials.

13. generating the set of chemical reaction schemas, accessing a chemical reaction inventory, the chemical reaction inventory including representations of chemical reactions describing the transformation of the raw materials into target products of the one or more target products; and populating the set of chemical reaction schemas from the chemical reaction inventory.

14. generating the set of chemical reaction schemas, simulating a first constituent reaction of a chemical reaction scheme of the set of chemical reaction schemes using a machine learning model; estimating an output of a reward function, wherein an output of the machine learning model serves as an input to the reward function; estimating a maximum value of the reward function by modifying an input to the machine learning model, wherein the input is an output from a second constituent reaction that precedes the first constituent reaction in the chemical reaction scheme.

15. 1. A computer-readable medium storing computer-executable instructions, the computer-executable instructions, when executed by one or more processors of a computer system, causing the computer system to: accessing characterization data for the raw material, the characterization data including one or more spectra collected according to one or more spectroscopic methods; using the characterization data to predict a set of constituent materials contained in the feedstock; predicting a material composition of the feedstock using the predicted set of constituent materials; identifying one or more target products using, at least in part, the predicted material composition of the raw materials; generating a set of chemical reaction schemas that enable the conversion of at least a portion of said feedstock into said one or more target products; and storing in a data store an identification of the material composition of the feedstock, the one or more target products, and the set of chemical reaction schemas.

16. The computer-executable instructions, when executed by one or more processors of a computer system, cause the system to: identifying one or more inputs to a fitness function, the one or more inputs describing a chemical reaction schema of the set of chemical reaction schemas; using the one or more inputs to generate an output of the fitness function; 16. The computer-readable medium of claim 15, further comprising: selecting an implementation scheme from the set of chemical reaction schemata according to the fitness function, the one or more inputs, and the one or more target products.

17. identifying the one or more target products accessing inventory information describing a set of products; and using the inventory information to identify an incomplete subset of the set of products as the one or more target products.

18. The inventory information is the amount of said feedstock available for conversion; the quality of said raw materials available for conversion; market data for said raw materials available for conversion; a quantity of the one or more target products available in a geographic region; the quality of the target products among the one or more target products available in a geographic region; or and market data for target products of the one or more target products available in a geographic region.

19. predicting the set of constituent materials contained in the raw material, accessing a library of spectra and associated metadata corresponding to the one or more spectroscopic techniques; identifying bands within the one or more spectra of the characterization data; and matching the bands to spectra of the library of spectra to predict constituent materials of the set of constituent materials.

20. generating the set of chemical reaction schemas, simulating a first constituent reaction of a chemical reaction scheme of the set of chemical reaction schemes using a machine learning model; estimating an output of a reward function, wherein an output of the machine learning model serves as an input to the reward function; 20. The computer-readable medium of claim 15, comprising: estimating a maximum value of the reward function by modifying an input to the machine learning model, wherein the input is an output from a second constituent reaction that precedes the first constituent reaction in the chemical reaction scheme.

Images