Automatic prediction of the usability of concrete for at least one intended use on a construction site

JP7596299B2Active Publication Date: 2024-12-09PERI GMBH
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2021560490
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-04-03
Filing Date
2020-03-20
Publication Date
2024-12-09
Estimated Expiration
2040-03-20

Smart Images

  • Figure 0007596299000001
    Figure 0007596299000001
  • Figure 0007596299000002
    Figure 0007596299000002
  • Figure 0007596299000003
    Figure 0007596299000003
Patent Text Reader

Abstract

A method (100) for training an artificial neural network (ANN) (1), wherein the ANN (1) predicts and / or classifies at least one quality measure (23a, 23b) for the usability of a concrete batch (2) at a construction site. 1. A method (200) for predicting and / or classifying the usability of a batch of concrete (2) at a construction site, the method (200) comprising: a step (210) in which a set of features (21) characterizing the material composition of the batch (2) is determined; a step (220) in which at least one measure (22) of the mechanical consistency of the batch (2) is determined; a step (230) in which the features (21) and the measure (22) of the mechanical consistency are provided as inputs (11) to a trained ANN (1); and a step (240) in which at least one prediction and / or classification (23a*, 23b*) of quality measures (23a, 23b) of the usability of the batch (2) for at least one intended use at the construction site is obtained as output (13) from the ANN (1). A method (300) for tracking the use of concrete batches (2) using a blockchain (4) and a smart contract (5) running on the blockchain (4).
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to the evaluation and further processing of externally delivered concrete batches at a construction site.

[0002] REFERENCE TO RELATED APPLICATIONS This application claims priority to German Patent Application No. 102019108779.1, filed on April 3, 2019, the entire scope of which is incorporated herein by reference.

[0003] Prior Art The processing of concrete on a construction site can only be planned in advance to a certain extent. A construction site is always, in a sense, a batch size of one, i.e. a one-off production. For example, the requirements for the concrete to be used depend in many ways on the conditions of the concrete formwork. Also, the properties of concrete will not always be the same everywhere, even if a predefined recipe is followed exactly, because the physical properties of sand and gravel, for example, vary from one natural reservoir to another, from which these natural raw materials are obtained.

[0004] Thus, situations may arise where a concrete batch that has been specifically delivered is not suitable for its specifically intended use at the construction site, which requires a quick solution as to what specifically should be done with that batch, leading to delays as one has to wait for usable concrete at the site that has already been formed.

[0005] Challenges and Solutions It is therefore an object of the present invention to make the usefulness of a concrete batch of concrete better predictable.

[0006] The above mentioned object is achieved according to the invention by a method for training an artificial neural network according to the independent claims, a method for predicting and / or classifying the usefulness of concrete batches according to another independent claim and a method for tracking the use of concrete batches according to a further independent claim. Further advantageous embodiments emerge from the dependent claims which refer to these independent claims.

[0007] Disclosure of the Invention Within the framework of the present invention, a method has been developed for training an artificial neural network ANN, which predicts and / or classifies at least one quality measure for the usability of a concrete batch on a construction site. The behavior of the ANN is characterized by a set of parameters, which may include, for example, weights, with which inputs for activating the neurons are calculated, which are respectively fed to the neurons and / or other computational units.

[0008] The method includes providing a series of training data sets, each of which includes a set of features for a batch of concrete that characterize the material composition of the batch, which may include, for example, the types and mixture ratios of the components of the concrete.

[0009] The training data set further includes at least one measure of the mechanical consistency of the batch, such as flowability, which can be determined, for example, using a standardized spreadability test.

[0010] The learning data set further comprises at least one value for a quality measure characterizing the usefulness of the batch for at least one intended use on the construction site. This quality measure may have been empirically determined by a human expert, for example, for the respective concrete batch. The quality measure may be binary in the simplest case (approval or rejection for the respective intended use), but may also, for example, indicate a scalar quality indicator related to the respective intended use. A single ANN may, inter alia, simultaneously provide quality measures for several different intended uses on the construction site, which are, for example, combined into one vector. In particular, for example, the time-temperature behavior of the hardening of the concrete batch may have a decisive influence on the quality measure. For example, it may be asked how long it took for the concrete to harden at a given temperature to such an extent that it is possible to remove the formwork from the concrete and use the formwork elements for the next cycle of formwork. The recognition of patterns in multiple states of the concrete batch may, for example, be used to improve the quality of the concrete.

[0011] During training, for each training data set, the set of features contained in the training data set and the measure of dynamic consistency contained in the training data set are respectively fed as input to the ANN, which produces as output a prediction and / or classification for at least one quality measure. This prediction and / or classification for the quality measure is compared with the value for the quality measure contained in the training data set. A cost function is used that is dependent on the deviation Δ determined in the comparison. The parameters of the ANN and / or the training data set are adapted with the optimization goal of improving the value of the cost function. The adaptation of the training data set may for example include underestimating or discarding training data sets that have errors or uncertainties in the data, so that such training data sets do not falsify the training results.

[0012] The parameters of the ANN evolve in the course of such training, starting for example from random starting values, so that the ANN predicts more or less well the quality measures of usability contained in the training data set together with the respective measures of mechanical consistency for the respective features contained in the training data set. This means that it is possible to use the experience of usability of several different concrete batches collected in the training data set to output an accurate prediction of the usability of future batches. If the set of training data sets is sufficiently large and diverse, the ANN can generalize the knowledge contained in the training data set and subsequently predict with sufficient accuracy the quality measures of usability of the relevant concrete batch for a specific purpose of use on a construction site for completely unknown combinations of features and mechanical consistency. It is precisely this generalization ability of the ANN that is important in the context of the evaluation of concrete. As mentioned at the beginning, concrete is produced from natural resources obtained from local stockpiles of the respective suppliers. When a similar building is constructed at a distant location using the same formwork, it is highly likely that the sand and gravel, for example, will come from other natural reservoirs and therefore have different physical properties than at the original construction site.

[0013] A similar impact can occur, for example, during a transition from a first construction site to a second nearby site, if the concrete supplier has to be changed, for example because the previous supplier is currently busy. The different suppliers essentially do not share their raw material sources with each other.

[0014] Furthermore, differences between different concrete batches can manifest themselves in different strengths, for example depending on temperature, so that, for example, a change in material composition in mid-summer may have little effect on the flowability of the concrete, whereas the same change in concrete in winter may make it much more viscous.

[0015] Thus, in a further particularly advantageous embodiment, the training data set comprises: · Identification of the location where at least one raw material used for the concrete batch was obtained; and / or Identification of the concrete batch supplier, and / or A measure of the ambient temperature at the time the quality measure is defined, and / or Information about what is being built on the construction site, and / or Information about where the concrete batches are to be delivered, and / or planning data, at least in excerpt form, relating to the building to be constructed, and / or At least an excerpt from the Building Information Model (BIM) of the building to be constructed, and / or Information regarding the origin, type and / or consistency of at least one component of the concrete batch; and / or Information regarding the proportion of at least one component of the concrete batch that is naturally occurring material and the proportion of that component that is recycled material; as further inputs to be fed to the ANN.

[0016] The BIM should be seen as a "digital twin" of the building to be constructed and can, for example, contain information about which concrete quality should be used at which site in addition to the building's geometric structure. The consistency of at least one component and also the consistency of the concrete batch as a whole can be measured, for example, using a spreadability test. Information about the extent to which at least one component of the concrete batch is recycled material can be used, for example, to find out what proportion of naturally obtained materials can be replaced by recycled material, with the aim of saving resources, without having to worry about the quality of the concrete as a whole.

[0017] In a particularly advantageous embodiment, the ANN to be trained is configured to additionally predict and / or classify at least one climatic quantity of the concrete batch, the climatic quantity being a characteristic quantity for climatic influences attributable to the concrete batch.

[0018] Correspondingly, at least some of the training data sets, preferably all of them, also contain values ​​of the climatic quantities for the respective concrete batches to which they relate. During training, the ANN additionally determines predictions for the climatic quantities from the features in the training data sets and the measures for the mechanical consistency. This prediction is compared with the values ​​for the climatic quantities contained in the respective training data sets. The cost function used for training therefore also depends on the deviation Δ′ determined in this comparison.

[0019] In this way, the trained ANN will also be able to quantify the climate impact of each concrete batch. This will allow, for example, to determine with greater precision the climate impact of an entire building constructed from many concrete batches. In different countries and regions, official management systems are being discussed or developed, according to which every citizen and company that causes a climate impact through a certain activity will be liable for a tax or levy graded according to this impact and / or will have to purchase the corresponding pollution rights in a national or international emission trading system. Thus, in the future, the total cost of constructing a building will depend heavily on the climate impact of the concrete used. If environmental laws become more stringent, building projects will have to be stopped due to excessive climate impacts, or may not be started at all in the first place.

[0020] Already today, the prospects of climate impacts are an important factor in tender competitions for the construction of buildings and for the environmental certification of buildings. For example, public financial support for the construction of a building may be linked to the condition that the building meets criteria for environmental certification. Building permits may also be linked to such conditions.

[0021] The climatic quantities may include, inter alia, a measure for the amount of at least one greenhouse gas that is emitted by and / or is bound to the concrete batch due to its production and / or use. Greenhouse gases here include, inter alia, CO2, which is generated during the production and processing of cement, for example. 2 Climatic quantities can also aggregate different types of climatic influences, for example according to a weighting or point system.

[0022] The trained ANN is then used in a method for predicting and / or classifying the usefulness of concrete batches at a construction site. As mentioned above, the ANN only needs to be trained once and can then be used in a large number of completely unknown situations. This so-called inference from a trained ANN requires significantly less computational power than training and can therefore be well performed, for example, by mobile equipment available at the construction site.

[0023] A set of features characterizing the material composition of the batch is determined for predicting and / or classifying the usefulness. Additionally, at least one measure of the mechanical consistency of the batch is determined. This determination can be based on manufacturer information, on measurements (e.g., using a spread test), or on any combination of manufacturer information and measurements, respectively.

[0024] The features and the measure of mechanical consistency are provided as input to a trained ANN, and at least one prediction and / or classification of a quality measure of the usefulness of the batch for at least one intended use on the construction site is obtained as output from the ANN.

[0025] Similarly as above, from the trained ANN, at least one prediction and / or classification of a climatic quantity that is a feature for the climatic impact attributable to the concrete batch can additionally be obtained as output, which climatic quantity can include, among others, for example, a measure for the amount of at least one greenhouse gas emitted due to the production and / or use of the concrete batch and / or that is bound to the concrete batch.

[0026] As above, the ANN has · Identification of the location where at least one raw material used for the concrete batch was obtained; and / or Identification of the concrete batch supplier, and / or Measures for ambient temperature at construction sites can be additionally provided as inputs, thereby taking into account the impact of these influence quantities on the usefulness.

[0027] In a further particularly advantageous embodiment, in response to the predicted quality measure for the first intended use not meeting a predetermined quality criterion, a further prediction and / or classification for the second intended use is obtained as output from the trained ANN. As mentioned above, the same ANN can provide predictions and / or classifications for multiple quality measures for multiple different intended uses at once, for example combined into one vector. In other words, by evaluating the same training data pool of the ANN, one or more predictions for very different intended uses can be provided.

[0028] In this way, it is possible to have a rapid alternative ready for use, for example in the surprising situation mentioned at the beginning, where the concrete batch turns out to be unsuitable for the first intended use, despite nominal compliance with the given recipe. Since the concrete, once mixed, must be constantly moved and begins to harden as soon as the movement stops, there is basically not much time available to search for alternatives. The automatic search via the ANN ensures, among other things, that the concrete batch is supplied to a high-value and still usable use. This reduces the probability that the concrete batch cannot be used at all on the construction site and must be disposed of as waste.

[0029] In a further advantageous embodiment, if the predicted quality measure for a certain use meets a predefined quality criterion, a means for supplying the batch to said use can be activated. For example, at least one dispensing or pumping device for the concrete can be activated so as to direct the batch to the location where it is needed according to the use. For example, electronic instructions can be sent to a concrete transport vehicle, controlled by a human or automatically, to move to the intended use location for the batch on the construction site and to fill a predefined space with the batch.

[0030] In a further particularly advantageous embodiment, the quality criterion can additionally depend on predicted climatic quantities for the concrete batch, which can be arbitrarily weighted with respect to the original quality measure depending on how significant the climatic influences are in the context of the overall construction project.

[0031] In this case, the climatic impacts, quantifiable by climatic quantities, and the quality of the concrete, quantifiable by quality measures related to the intended use, can for example be non-linearly coupled to each other. Thus, for example, an improvement in the quality of the concrete, such that its quality measure increases from 80% to 100%, may require energy and material inputs that increase the climatic impacts by a factor of five.

[0032] This in itself can already have an impact on the planning stage of a construction project: narrow concrete structures that can be made with a small amount of concrete and therefore require concrete of 100% quality and the best ecological "footprint" can be penalized so highly by the above mentioned management system that bulkier structures that have twice the amount of concrete but therefore only need 80% of the quality are more economical.

[0033] The invention also relates to a method, closely related to the above method, for tracking the usage of concrete batches, in which a set of features characterizing the material composition of the batch and / or one or more hash values ​​formed from these features are associated with the batch and stored in a blockchain.

[0034] A blockchain in this context is a distributed data memory in which each block containing a payload additionally contains one or more hash values ​​of the previous block. In this way, the blocks are chained together, making it impossible to silently change the payload in one of the blocks in the chain, unless all subsequent blocks in the chain are adapted accordingly. However, blockchains, especially in decentralized networks such as the Internet, are often managed in such a way that several participants of the network have to "apply" to be able to add the next block to the blockchain by solving complex computational tasks. Thus, later changes to the history often require costs that are no longer technically and economically affordable. Prominent examples of blockchains are the Bitcoin blockchain, which acts as a global "registry" for all bitcoins existing in the world, and the Ethereum blockchain. The latter Ethereum blockchain, like the Bitcoin blockchain, not only acts as a "registry" for all units of the cryptocurrency ether, but is also explicitly designed to store any kind of payload and to automatically execute "smart contracts".

[0035] So the most important feature of a truly public blockchain like the Bitcoin blockchain or the Ethereum blockchain is that data can be stored in the blockchain in such a way that it cannot be lost and cannot be changed later. At the same time, the data is publicly available for everyone. If this is not desirable, private blockchain networks can also be used.

[0036] By storing the hash values ​​of the features that characterize the material composition, the exact material composition can be kept secret, for example when using a public blockchain.

[0037] Storing "associated with a batch" means that for a given batch, there is a path to retrieve from the blockchain the information that exactly belongs to this batch. For this purpose, the information can be stored in the blockchain, for example in combination with an identifying characteristic of the batch.

[0038] Features characterizing the material composition of a batch, or a hash value thereof, can be stored in the blockchain, for example by the manufacturer of the batch, e.g. by a cement factory.

[0039] Furthermore, a measure of the mechanical consistency of the batch is physically determined, for example using a spread test, and stored in the blockchain in association with the batch, which can be performed, for example, by the recipient of the concrete at the construction site, which is meaningful as the properties of the concrete may have further changed during transport from the cement plant to the construction site.

[0040] Furthermore, a quality measure for the usefulness of the batch for at least one intended use at the construction site is determined and associated with the batch and stored in the blockchain, which can also be performed, for example, by the recipient of the concrete at the construction site.

[0041] Storing information in a blockchain offers the particular advantage in this context of making it possible to merge information from several different sources and at the same time protecting it from subsequent manipulation without the need for passwords or other access data to be centrally managed. Protection against manipulation is important, since the data may be needed later for evidentiary purposes. For example, a complaint about lack of quality against a concrete manufacturer can be used later to investigate the background of the stability of buildings manufactured with this concrete. Similarly, for example, a claim can be made later on that a poor quality concrete that can only be used for relatively low-level purposes was delivered multiple times during a given period, to request a partial refund of the concrete price from the manufacturer.

[0042] In a particularly advantageous embodiment, at least one climatic quantity is additionally determined, which is characteristic of the climatic impact attributable to the concrete batch, and stored in the blockchain in association with the batch. This climatic quantity can be determined, for example, by the above-mentioned method, but also by any other method. For example, the tamper-proof storage in the blockchain allows evidentially and permanently documenting the complete ecological "footprint" of the building.

[0043] In a particularly advantageous embodiment, the quality measure is obtained as a prediction and / or classification from the trained ANN by the above method and / or is at least plausibly checked by the prediction and / or classification thus obtained. For example, it is possible to refuse to store a very poor value for the quality measure in the blockchain if, based on the trained ANN, a very good value for the quality measure would be expected, other things being equal. In contrast, in this same situation, an average value for the quality measure is assumed, since the variation between "very good" and "medium" is reasonable based on the physical variations of the process.

[0044] In this way, it can at least be made more difficult to deliberately store a particularly poor assessment in the blockchain with fraudulent intent, which could for example be to blackmail the concrete producer or to subsequently demand from the latter a refund of part of the price paid, pointing out a falsely-phrased quality.

[0045] In a further particularly advantageous embodiment, the actual purpose for which the concrete batches are used at the construction site is stored in the blockchain in association with the batches. As mentioned above, this information can be related to a price reduction based on, for example, lack of quality. Thus, for example, it can be contractually stipulated that if a nominally high-quality batch can in fact only be used for less demanding purposes, then only the price for which such a relatively low-quality batch would have been sold should be paid.

[0046] In a further particularly advantageous embodiment, a smart contract running on the blockchain determines a price for the batch and credits the supplier of the batch according to predefined criteria based on data associated with the batch and stored in the blockchain.

[0047] A smart contract, in this context, is a program that runs synchronously on all nodes participating in the blockchain, i.e. that performs the same operations. A smart contract can, among other things, access all data stored in the blockchain, but can also register new information to add to the blockchain. Within the blockchain, a smart contract can be managed as an entity that can receive and independently manage funds, among other things in the form of the cryptocurrency on which the blockchain is based. Thus, for example, a smart contract can first receive a maximum purchase price from a buyer of concrete and later transfer the set price to the supplier, while the buyer is refunded the remaining amount.

[0048] In this way, price changes, especially due to quality deficiencies, can be planned and automatically implemented according to clearly traceable objective criteria, and correspondingly time-consuming discussions are avoided.

[0049] Thus, advantageously, the predefined criteria depend at least on a quality measure for the usefulness of the batch for at least one intended use on the construction site and / or on the actual intended use for which the batch is used on the construction site.

[0050] Likewise, the criteria previously defined in the framework of the smart contract can also depend on the climatic quantities for the batch, which can be obtained, inter alia, for example, from the blockchain, but also from any other source. For example, taxes due for climatic impacts can be deducted from the amount to which the taxpayer is entitled and paid to the competent authority at the time of their occurrence based on the decisions of the aforementioned management system.

[0051] Criteria once stored in a smart contract essentially cannot be changed later: any mutability here would have to be explicitly contemplated from the start in the program code of the smart contract and would be visible there.

[0052] The provision and automatic enforcement of contract terms using smart contracts can be extended to further actors related to the construction site. Thus, for example, in addition to the construction site itself and the cement plant, planning offices, general contractors, and formwork manufacturers are often also involved. For example, the overall plan of a construction project can be stored in the blockchain. The progress of the construction can be detected electronically using any indicators at the construction site and stored in the blockchain as well. The smart contract can automatically order the payment of installments to certain actors, for example, when certain milestones defined in the overall plan are reached.

[0053] In this way, the entire construction project can be better protected from any of the parties being deceived. For example, the smart contract can stipulate that at the beginning of the construction project, the building owner must deposit some or all of the construction costs into the smart contract for trusteeship. This ensures that all actors providing services in relation to the construction project will receive remuneration from liquid funds after providing their respective services. There is no longer a risk, as is common with open accounts at the time of delivery and services, that the building owner will have difficulty paying and the actors who provided the services in advance will experience liquidity problems while collecting the outstanding amount through legal means.

[0054] On the contrary, the owner has the certainty that he will always be paid only for deliveries and services actually rendered. He must therefore deposit his money in trust with the smart contract, but this does not equate to having to make an advance payment to any actor, i.e., if one of the actors, for example the cement plant, goes bankrupt, the owner's capital will not be lost.

[0055] The trusteeship of funds by smart contracts can further preclude any person from gaining direct access to these funds and having the opportunity to embezzle them. Funds once deposited in a smart contract for trusteeship can only be returned under conditions stipulated in the smart contract, such as, for example, when a certain milestone is reached. The smart contract can also include, for example, a termination condition, such as a construction project being officially declared a failure if there is no construction progress for a certain period of time, or the owner of the building having the money deposited in the smart contract returned after all actors have been remunerated for the services already rendered.

[0056] For example, a delivery note can also be stored in the blockchain, which is then stored in such a way that it cannot be lost and is protected against manipulation.

[0057] The data stored in the blockchain can be retrieved, for example, at the concrete manufacturer and used there to improve the concrete quality. Furthermore, by storing the data persistently in the blockchain, it is possible to document the compliance, which is important in terms of stability, with the specifications of the concrete used, even over a period of the order of the life cycle of the building. It is thus possible to later order the inspection of parts of the building or to shorten the periodic inspection deadlines associated therewith, in response to the exposure of defects in the data from the construction period, for example.

[0058] The invention also relates to an artificial neural network trained by the above training method. The invention also relates to a data set of parameters characterizing the ANN obtained by the above training method. The highly computationally intensive training, which often requires, for example, a GPU with an extraordinarily large video memory, can be offered as a service and the data set can be delivered as a work result of this service.

[0059] The present invention also relates to one or more computer programs having machine-readable instructions that, when executed on one or more computers and / or on a blockchain, cause the one or more computers and / or blockchain to perform one or more of the above methods. For example, a program running on a blockchain can cooperate with other programs that send or receive information to or from the blockchain.

[0060] The computer program or programs may, for example, be embodied on a machine-readable data carrier or on a download product obtainable and loadable via a network.

[0061] Detailed explanation section The subject matter of the present invention will be described below with reference to the drawings, but the subject matter of the present invention is not limited to the drawings. [Brief description of the drawings]

[0062] [Figure 1] FIG. 1 illustrates an embodiment of a method 100 for training an ANN1. [Diagram 2] FIG. 2 illustrates an example embodiment of a method 200 for predicting and / or classifying the utility of a concrete batch 2. [Diagram 3] FIG. 1 illustrates an embodiment of a method 300 for tracking usage of a concrete batch 2. [Figure 4]FIG. 2 illustrates an example of the application of the method 200 on a construction site.

[0063] 1 shows an exemplary flow chart of a method 100 for training an ANN 1. In step 110, a training data set 3 is prepared to "feed" the ANN. In the example shown in FIG. 1, each training data set for a specific concrete batch 2 includes: A set of features 21 characterizing the material composition of batch 2, e.g., weight ratios of the components; a measure 22 of the mechanical consistency of batch 2, determined for example using a spreading test; at least one value for a quality measure 23a, 23b characterizing the usefulness of the batch for at least one purpose on the construction site and set, for example, by a human expert who has received the batch 2; an identifier 24 of the location where the raw materials used for batch 2 were obtained; - Batch 2 supplier identifier 25; Scale 26 for the ambient temperature at the time when quality scales 23a, 23b are specified; Includes.

[0064] In step 120, all this information except the quality measures 23a, 23b is fed as input 11 to the ANN 1. The ANN 1 then provides a prediction and / or classification 23a*, 23b* for the quality measures 23a, 23b. In step 130, this prediction and / or classification 23a*, 23b* is compared with the values ​​for the quality measures 23a, 23b contained in the training data set. The comparison provides a deviation Δ.

[0065] In step 140, a cost function ("loss function") 14, which is dependent on the deviation Δ, is evaluated. In step 150, the parameters 12 of the ANN are adapted with the optimization goal of improving the value of the cost function 14. For example, the parameters 12 can be successively adapted by gradient descent until the average value of the cost function 14 obtained over all training data sets 3 falls below a predefined threshold.

[0066] According to block 121, the ANN 1 also provides a prediction 23c* for the climatic quantity 23c. According to block 131, this prediction 23c* is compared with the values ​​for the climatic quantity 23c contained in the respective training data set 3. Then, according to block 141, the cost function 14 is additionally dependent on the deviation Δ′ determined in the comparison 131.

[0067] 2 shows an example embodiment of a method 200 for predicting and / or classifying the usefulness of a concrete batch 2 at a construction site. The method 200 assumes the availability of an ANN that has been trained, for example, by the method 100 described above.

[0068] In step 210 a set of features 21 characterizing the material composition of batch 2 is determined. In step 220 at least one measure of the mechanical consistency of batch 2 is determined, for example using a spread test. Additionally, an identifier 24 of the location where the raw materials for batch 2 were obtained, an identifier 25 of the manufacturer of batch 2 and a measure 26 of the ambient temperature at the construction site may also be used.

[0069] The collected information is fed as input 11 to the ANN1 in step 230. The ANN1 then outputs a prediction and / or classification 23a*, 23b* for at least one quality measure 23a, 23b related to the usefulness of the batch 2 for at least one intended use on the construction site. The prediction and / or classification 23a*, 23b* is thus always relevant to the intended use at the time. This prediction and / or classification 23a*, 23b* is obtained from the ANN1 in step 240.

[0070] In this case, it may occur inter alia that the first prediction and / or classification 23a* indicates that the batch 2 is probably unsuitable for the first intended use. In this case, according to block 241, a second prediction and / or classification 23b* can be obtained which is associated with a quality measure 23b of usability for another intended use. For example, a batch 2 which is not suitable for manufacturing a particularly complex and at the same time highly loaded structure of a building or part of a building to be constructed can be used for concreting a less important structure.

[0071] According to block 242, at least one prediction and / or classification 23c* of climatic quantities 23c, which are features for the climatic influences attributable to batch 2, can additionally be obtained as output 13 of the ANN 1.

[0072] In step 250, means can be activated for delivering batch 2 to the intended use that has been deemed suitable according to the predictions and / or classifications 23a*, 23b*.

[0073] The corresponding quality criteria may additionally depend on the climatic quantities 23c* predicted for batch 2. As mentioned above, concretes may be optimal which, for example, only slightly exceed the minimum requirements rather than the maximum requirements with regard to classification 23a*, but therefore have a significantly reduced ecological "footprint".

[0074] 3 shows an exemplary flow chart of a method 300 for tracking the usage of a batch of concrete 2. In step 310, features 21 characterizing the material composition of the batch 2 are stored in association with this batch 2 in the blockchain 4. In step 320, a measure 22 for the mechanical consistency of the batch 2 is physically determined, for example using a standardized spreading test. The result is stored in association with the batch 2 in the blockchain 4 in step 330.

[0075] In step 340, a quality measure 23a, 23b for the usefulness of the batch 2 for at least one intended use on the construction site is determined. This quality measure 23a, 23b is stored in the blockchain 4 in association with the batch 2 in step 350.

[0076] According to block 341, the quality measures 23a, 23b can be determined as predictions and / or classifications 23a*, 23b* according to the above method 200. According to block 342, the quality measures 23a, 23b can be plausibly checked by such predictions and / or classifications 23a*, 23b*, for example to prevent unjustified and overly negative evaluations of batch 2.

[0077] According to block 343, at least one climatic quantity 23c that is a feature for the climatic impact attributable to batch 2 can also be determined and, according to block 353, can be associated with batch 2 and stored in the blockchain 4.

[0078] Further, in step 335, an identifier 24 of the location where the raw materials for batch 2 were obtained, and / or an identifier 25 of the manufacturer of batch 2, and / or a measure 26 of the ambient temperature at the time when the quality measures 23a, 23b of the usability of batch 2 were determined, may also be stored in the blockchain 4.

[0079] In step 360 , the actual use 27 for which the concrete batch 2 is to be used at the construction site is associated with the batch 2 and stored in the blockchain 4 .

[0080] In step 370, the smart contract 5 running on the blockchain 4 determines a price 2a for batch 2 based on all data 22, 23a, 23b, 24, 25, 26, 27 previously associated with batch 2 and stored in the blockchain 4. The criteria 5a applied here are hardwired, so to speak "soldered" into the smart contract 5 and cannot be changed later. In step 380, the smart contract 5 credits this price 2a to the supplier of batch 2.

[0081] Figure 4 shows a concrete example of how the use of a concrete batch 2 at a construction site can be controlled by the method 200. In this example, the batch 2 is mixed from four components: sand 6a, gravel 6b, cement 6c, and water 6d. Feature 21 indicates the mixing ratio of these components 6a-6d. The sand 6a and gravel 6b come from a natural reservoir at location 24. The batch 2 is manufactured by a manufacturer 25.

[0082] Here, first of all a spreading test is carried out on the construction site in order to determine a measure 22 for the mechanical consistency of batch 2. This measure 22 is fed to the trained ANN 1 together with the features 21, the location identifier 24, the manufacturer identifier 25 and the measure 26 for the temperature at the construction site.

[0083] Based on this information, the ANN 1 determines a first prediction and / or classification 23a* for a quality measure 23 related to a first intended use, here the concrete processing of the currently formed part 7a' of the concrete arch 7a. If the quality of batch 2 according to the prediction and / or classification 23a* is sufficient for this intended use (truth value 1), batch 2 is used for this intended use.

[0084] If, on the other hand, the quality of batch 2 is not sufficient for this intended use (truth value 0), then a second prediction and / or classification 23b* is obtained for a less demanding intended use of batch 2, here a quality measure 23b relating to concrete processing for construction roads 7b. If, according to the prediction and / or classification 23b*, batch 2 is suitable for the latter intended use (truth value 1), batch 2 is used accordingly. If, on the other hand, batch 2 is not suitable for the latter, less demanding intended use either (truth value 0), then batch 2 is discarded as waste. [Explanation of symbols]

[0085] 1. Artificial Neural Network (ANN) 11 Input of ANN1 12 ANN1 parameters 13 Output of ANN1 14 Cost function for training ANN1 2. Concrete batch 2a Price for batch 2 21 Features characterizing the composition of batch 2 22 Measures of mechanical consistency for batch 2 23a Quality measures of batch 2's usefulness for the first objective 23a* Prediction / classification of quality measures 23a provided by ANN1 23b Quality measures of batch 2's usefulness for the second objective 23b* Prediction / classification of quality measures 23b provided by ANN1 23c Climate quantities that are measures of climate impacts in batch 2 23c* Prediction / classification of climatological quantities 23c provided by ANN1 24 Identifier of the location where the raw materials for batch 2 were obtained 25 Manufacturer's Identifier for Batch 2 26 Temperature scale 27 Practical use of batch 2 3. Learning dataset for training ANN1 4. Blockchain 5 Smart contracts that run on blockchain 4 5a Criteria for Pricing in Smart Contracts 5 6a Sand 6b Gravel 6c Cement 6d water 7a Concrete Arch 7a' Part of concrete arch 7a with the formwork removed 7b construction road 100 Methods for training ANN1 110 Prepare training dataset 3 120 Feed input 11 to ANN1 121 Prediction / Classification by ANN1 23c* is also provided 130 Comparing predictions / classifications 23a*, 23b* with quality measures 23a, 23b 131 Comparing prediction / classification 23c* with climatological quantities 23c 140 Evaluate cost function 14 141 also evaluates the deviation Δ′ determined in comparison 131 150 Adapt parameter 12 of ANN1 200 Methods for predicting / classifying usefulness of batch 2 210 Determine feature 21 220 Determine a measure of the mechanical consistency of batch 2 230 Input 11 is fed to ANN1 240 Obtain prediction / classification 23a*, 23b* from ANN1 241 Obtain further predictions / classifications 23a*, 23b* 242 Also get prediction / classification of climatology 23c* 250 Drive means to supply batch 2 for use 300 A method for tracking batch 2 usage 310 Store the feature 21 or hash value in the blockchain 4 320 Determine a measure 22 of the mechanical consistency of batch 2 330 Store scale 22 in blockchain 4 335 Store further information 24, 25, 26 in the blockchain 4 340 Determine quality measures 23a, 23b 341 Determine quality measures 23a, 23b as prediction / classification 23a*, 23b* 342 Validating quality measures 23a, 23b by prediction / classification 23a*, 23b* 343 Determine climatological quantity 23c 350 Store quality measures 23a and 23b in the blockchain 4 353 Climate quantity 23c is also stored in blockchain 4 360 Actual usage 27 stored in blockchain 4 370 Determine price 2a based on criterion 5a 380 Credit price 2a to batch 2 supplier Δ deviation determined in comparison 130 Δ′ deviation determined in further comparison 131

Claims

1. A method (100) for training an artificial neural network (ANN) (1), implemented by a computer comprising said ANN (1), comprising: the ANN (1) functions to predict and / or classify, for a concrete batch (2), quality measures (23a, 23b) characterizing the suitability of the batch (2) for at least one intended use on a construction site, using as input a set of features (21) characterizing a material composition of the batch (2) and at least one measure (22) of the mechanical consistency of the batch (2), the behavior of the ANN (1) being characterized by a set of parameters (12); The method (100) comprises: - receiving (110) a set of training data sets (3), each comprising a set of features (21) characterizing the material composition of the batch (2), at least one measure (22) of the mechanical consistency of the batch (2), and at least one value of a quality measure (23a, 23b) characterizing the usability of the batch (2) for at least one intended use on the construction site; a step (120) of receiving, for each of the training data sets (3), a set of the features (21) contained in the training data set (3) and a measure (22) for the dynamic consistency contained in the training data set (3) as input (11), and outputting (13) a prediction and / or classification (23a*, 23b*) for the at least one quality measure (23a, 23b); a step (130) of comparing said predictions and / or classifications (23a*, 23b*) for said quality measures (23a, 23b) with values ​​for said quality measures (23a, 23b) contained in said training data set (3); - evaluating (140) a cost function (14) dependent on the deviation Δ determined in said comparison (130); - adapting (150) the parameters (12) of the ANN (1) with the optimization goal of improving the value of the cost function (14); having the ANN (1) causes the computer to additionally predict and / or classify at least one climatic quantity (23c) that is a characteristic of a climatic impact attributable to the concrete batch (2); the training data set (3) also comprises values ​​of the climatic quantities (23c) for each of the concrete batches (2) to which the training data set (3) relates, The ANN (1) additionally determines (121) a prediction (23c*) for the climatic quantity (23c) from the features (21) of the training data set (3) and the measure (22) of dynamic consistency, comparing (131) the predictions (23c*) for the climatic quantities (23c) with the values ​​for the climatic quantities (23c) contained in the respective training data set (3); the cost function (14) additionally depends (141) on the deviation Δ′ determined in the comparison (131), Method (100).

2. the climatic quantity (23c) comprising a measure for the amount of at least one greenhouse gas emitted by the production and / or use of the concrete batch (2) and / or bound to the concrete batch (2), The method of claim 1.

3. The learning data set (3) is an identifier (24) of the location where at least one raw material used for the concrete batch (2) was obtained, and / or - an identifier (25) of the supplier of the concrete batch (2), and / or a measure (26) for the ambient temperature at the time when said quality measures (23a, 23b) are defined, and / or - information about what is being built at the construction site, and / or information regarding where the concrete batch (2) is to be delivered, and / or - planning data, at least in excerpt form, relating to the building to be constructed, and / or At least an excerpt from a Building Information Model (BIM) of the building to be constructed, and / or - information regarding the origin, type and / or consistency of at least one component of the concrete batch (2), and / or Information regarding what proportion of at least one component of the concrete batch is naturally occurring material and what proportion of that component is recycled material. as further inputs (11) to be fed to said ANN (1), The method (100) of claim 1 or 2.

4. An artificial neural network ANN (1) trained by the method (100) according to any one of claims 1 to 3.

5. A computer-implemented method (200) for predicting and / or classifying the usefulness of a concrete batch (2) at a construction site, comprising: The method (200) comprises: - determining (210) a set of features (21) characterizing the material composition of said batch (2); - determining (220) at least one measure (22) of the mechanical consistency of said batch (2); - providing (230) said features (21) and said measure of dynamic consistency (22) as inputs (11) to a trained ANN (1); obtaining (240) as output (13) from the ANN (1) at least one prediction and / or classification (23a*, 23b*) of a quality measure (23a, 23b) of the usefulness of the batch (2) for at least one intended use on the construction site; having The method (200) additionally obtains (242) from the trained ANN (1) at least one prediction and / or classification (23c*) of a climatic quantity (23c) that is a feature for a climatic impact attributable to the concrete batch (2) as an output (13).

6. the climatic quantity (23c) comprising a measure for the amount of at least one greenhouse gas emitted by the production and / or use of the concrete batch (2) and / or bound to the concrete batch (2), The method of claim 5.

7. The ANN (1) includes: an identifier (24) of the location where at least one raw material used for the concrete batch (2) was obtained, and / or - an identifier (25) of the supplier of the concrete batch (2), and / or A measure for the ambient temperature at the construction site (26) is additionally provided as input (11), The method (200) of claim 5 or 6.

8. In response to the predicted quality measure (23a*) for the first intended use not meeting a predefined quality criterion, a further prediction and / or classification (23b*) for a second intended use is obtained (241) as output from the trained ANN (1); The method (200) of any one of claims 5 to 7.

9. In response to the predicted quality measures (23a*, 23b*) for a given intended use satisfying a predetermined quality standard, a means for supplying the batch (2) to said intended use is activated (250); The method (200) of any one of claims 5 to 8.

10. the predetermined quality criterion is additionally dependent on predicted climatic quantities (23c*) for the concrete batch (2), The method (200) of claim 9.

11. 1. A computer-implemented method (300) for tracking usage of a concrete batch (2), comprising: The method (300) comprises: storing (310) in the blockchain (4) a set of features (21) characterizing the material composition of said batch (2) and / or one or more hash values ​​formed from said features (21) in association with said batch (2); - physically determining (320) and storing (330) in the block chain (4) a measure (22) of the mechanical consistency of the batch (2); determining (340) a quality measure (23a, 23b) for the usefulness of the batch (2) for at least one intended use on a construction site based on the features (21) and the mechanical consistency and storing (350) it in the block chain (4) in association with the batch (2); having additionally determining (343) at least one climatic quantity (23c) which is a characteristic of a climatic impact attributable to said concrete batch (2) and storing (353) said at least one climatic quantity in said blockchain (4) in association with said batch (2); The method (300) further comprises:

12. Additionally, an identifier (24) of the location where at least one raw material used for the concrete batch (2) was obtained, and / or - an identifier (25) of the supplier of the concrete batch (2), and / or a measure (26) for the ambient temperature at the time when said quality measures (23a, 23b) are defined is stored in the blockchain (4) in association with the batch (2) (335); The method of claim 11.

13. The quality measures (23a, 23b) are determined (341) as a prediction and / or classification (23a*, 23b*) by the method (200) according to any one of claims 5 to 10, and / or A validation check (342) is performed with the predictions and / or classifications (23a*, 23b*) obtained by the method (200) according to any one of claims 5 to 10, The method (300) of claim 11 or 12.

14. The actual purpose (27) for which the concrete batch (2) is used at the construction site is associated with the batch (2) and stored in the blockchain (4) (360). The method (300) of any one of claims 11 to 13.

15. A smart contract (5) running on the blockchain (4) determines (370) a price (2a) for the batch (2) according to predefined criteria (5a) based on data associated with the batch (2) and stored in the blockchain (4) and credits (380) the supplier of the batch (2); The method (300) of claim 12.

16. An actual purpose (27) for which the concrete batch (2) is used at the construction site is associated with the batch (2) and stored (360) in the blockchain (4); A smart contract (5) running on the blockchain (4) determines (370) a price (2a) for the batch (2) according to predefined criteria (5a) based on data associated with the batch (2) and stored in the blockchain (4), and credits (380) the supplier of the batch (2); The predefined criteria (5a) are at least and / or depending on said quality measures (23a, 23b) on the usefulness of said batch (2) for said at least one intended use on said construction site; and / or depending on the actual purpose for which the batch (2) is used at the construction site, The method (300) of claim 12.

17. The predefined criteria (5a) are additionally dependent on the climatic quantities (23c) for the batch (2), 20. The method (300) of claim 16.

18. One or more computer programs comprising machine-readable instructions that, when executed on one or more computers, cause the one or more computers to perform a method (100, 200, 300) according to any one of claims 1 to 3, 5 to 17.

19. 20. A machine-readable medium having one or more computer programs according to claim 18.

20. 20. A computer system loaded with one or more computer programs according to claim 18.

Citation Information

Patent Citations

  • A Concrete Strength Prediction Method Based on a Hybrid Model

    CN104991051B

  • Method and apparatus for quality control of concrete

    JP1995227832A

  • Estimating device of consistency of fresh concrete

    JP1999077659A

  • Mix proportion-setting device for concrete

    JP1999077660A

  • Computing unit for carbon-dioxide emissions caused by use of concrete, computing method for carbon-dioxide emissions, computing program for carbon-dioxide emissions and computer-readable recording medium

    JP2012159892A