System and method for optimizing the co 2 equivalent emissions of a machinery

Abstract

Described herein is a method, implemented through a computer (17), for measuring and reducing the carbon footprint emitted by machinery (3), comprising the steps of: receiving at least one variable v i associated with at least one actuator (5) of the machinery (3) and relating to a phase of a manufacturing process; reading, from a database (19), a plurality of computation parameters α i for determining the carbon footprint of the phase of the manufacturing process; computing the equivalent carbon footprint CO2eq of the machinery (3) on the basis of the at least one variable v i and the plurality of computation parameters α i ; reading, from memory means (21), a predetermined threshold value; when the value of the computed equivalent carbon footprint CO2eq is higher than the predetermined threshold value: - providing, through a machine learning algorithm trained by means of a regularized linear regression technique, at least one command for the at least one variable v i capable of reducing the value of the equivalent carbon footprint CO2eq; - generating a control signal (20) comprising suitable commands to be applied to the at least one variable v i for reducing the equivalent carbon footprint CO2eq of said machinery (3).

Description

[0001] SYSTEM AND METHOD FOR OPTIMIZING THE CO2EQUIVALENT

[0002] EMISSIONS OF A MACHINERY

[0003] DESCRIPTION

[0004] The present invention relates to a system and a method for optimizing equivalent CO2 emissions of machinery.

[0005] Equivalent CO2 is a unit of measurement which is used for evaluating the impact of various greenhouse gases (GHGs) on global warming.

[0006] Such unit of measurement expresses the amount of CO2 that would have the same global warming effect as the considered gases, computed according to the Global Warming Potential (GWP) index.

[0007] By using equivalent CO2 as a unit of measurement, it is possible to evaluate the global environmental impact of an activity, event or product, even when different types of greenhouse gases are emitted. Thus, all types of greenhouse gases can be considered in terms of equivalent amounts of carbon dioxide, thereby facilitating the evaluation and comparison of their impact on global warming.

[0008] Machinery, in particular industrial machinery, plays a fundamental role in the manufacturing field, allowing the production of essential goods and services. However, the use of machinery can lead to considerable emissions of equivalent CO2.

[0009] Industrial processes often require the use of energy coming from non-renewable sources, such as coal or natural gas. During production, such energy sources release carbon dioxide into the atmosphere, thus contributing to the greenhouse effect and to the climate change. Moreover, some industrial activities may also produce other greenhouse gases, such as methane or nitrogen oxide.

[0010] Systems and / or methods are known in the art for computing the carbon footprint of an industrial plant or industrial machinery.

[0011] For example, patent application no. WO 2023 / 285389 concerns, in general, measurement devices and control devices for optimizing environment-relevant energy consumption in a production process of an industrial plant or site, which can optimize both electric and thermal energy to obtain desired overall performance from such plant. However, while acknowledging that it is necessary to act upon the actuators of the machinery in order to be able to reduce the environmental footprint of the latter, it is very difficult to decide which actuator(s) should be considered, since the measured variables of the machinery are not necessarily independent of one another.

[0012] It is therefore the object of the present invention to provide a system and a method for optimizing equivalent CO2 emissions of machinery, taking into account the nature of the quantities contributing to such emissions.

[0013] Further advantageous features of the present invention are set out in the appended claims, which are an integral part of the present description.

[0014] These features as well as further advantages of the present invention will become more apparent in the light of the following description of a preferred embodiment thereof as shown in the annexed drawings, which are provided herein merely by way of nonlimiting example, wherein:

[0015] Figure 1 shows a system for optimizing equivalent CO2 emissions of machinery according to the invention;

[0016] Figure 2 shows a scheme of iterations of subsets of a dataset for training a supervised machine learning model used in the method according to the invention.

[0017] The following description will illustrate several specific details to facilitate an in-depth understanding of one or more exemplary embodiments of the invention. The various embodiments may also be implemented without one or more of such specific details or by using other methods, components, materials, etc. In some cases, known structures, materials or operations will not be shown or described in detail to avoid shadowing some aspects of such embodiments. In this description, any reference to an “embodiment” will indicate that a particular configuration, structure or feature described herein in relation to an embodiment is comprised in at least one embodiment of the invention. Therefore, expressions such as “in one embodiment” and the like, which may be found in different parts of this description, will not necessarily refer to the same embodiment. Moreover, any particular configuration, structure or feature may be combined as deemed appropriate in one or more embodiments.

[0018] The references below are therefore used only for simplicity’s sake and shall not limit the protection scope or extension of the various embodiments.

[0019] With reference to Figure 1, the following will describe a system 1 according to the invention for optimizing equivalent CO2 emissions of machinery.

[0020] The system 1 comprises machinery 3, e.g. industrial machinery, an edge device 7, or programmable logic controller, suitable for being associated with the machinery 3, and a computer 17, in particular a remote processing unit. The machinery 3 is suitable for executing a series of operations or processes within a production environment. It may comprise a plurality of actuators 5 that convert electric, pneumatic or hydraulic energy into mechanical motion. The actuators 5 may be electric, pneumatic or hydraulic, depending on specific requirements of the machinery.

[0021] The actuators 5 may perform a wide range of functions in the machinery 3. For example, actuators may be used for opening and closing valves, lifting and lowering loads, moving movable parts, adjusting the speed, and so forth.

[0022] The edge device 7 comprises, connected to its input ports, a plurality of sensors 9 suitable for detecting and measuring a plurality of variables w within the machinery 3, such as, for example, temperature, pressure, speed, fluid level, position, and other quantities. Such quantities may be acquired in an analogue or digital manner.

[0023] The edge device 7 further comprises a microprocessor 13 and memory means 15, e.g. a semiconductor memory.

[0024] The variables w collected by the one or more sensors 9 of the edge device 7 are continuously transmitted to the remote processing unit 17 in real time through connection means 16, e.g. a wired, wireless, LORA, Ethernet or LTE / 5G connection, to be then processed and used for making decisions about the management of the machinery 3, and particularly for preventing the environmental carbon footprint of the machinery 3 from exceeding a predetermined threshold and, should this happen, taking appropriate technical measures in order to adaptively bring such footprint back below the predetermined threshold.

[0025] The system 1 is configured to bring the carbon footprint value back below the predetermined threshold also by means of successive iterations progressively decrementing said footprint value.

[0026] The remote processing unit 17 comprises a microprocessor 18, associated with a computation module 23, and has access to a database 19.

[0027] The variables w are processed by the remote processing unit 17 through an algorithmic transformation based on the computation model described in document “Bern model (Bio-Environmental Regulator Model) - Global Warming Potentials (GWP) over a 100 year time horizon (based on IPCC 2013)”, as defined in the European Union’s Recommendation no. 2279 / 2021 on the use of environmental footprint methods to measure and communicate the life cycle environmental performance of products and organisations, available on the web site http s : / / eur-1 ex . euro a. eu / 1 egal - content / IT / TXT / PDF / ?uri=CELEX:32021H2279.

[0028] The remote processing unit 17 is in data communication with the database 19, in particular an LCI (“Life Cycle Inventory”), which contains a set of data describing the resource consumption, emissions, generated waste and energy used during the various phases of the life cycle of a product, service or process.

[0029] Such data may comprise parameters m like: raw materials used, comprising amounts and origins thereof; energy used, whether direct or indirect, during the various phases of the life cycle; atmospheric emissions, e.g. emissions of greenhouse gases and atmospheric pollutants; emissions of water, e.g. wastewater drains or hydric pollution; generation of waste, comprising amounts and composition thereof; impact on human health, e.g. impact of emissions on people; impact on the ecosystem, e.g. impact of emissions on the natural environment.

[0030] Based on the parameters cuand the variables w, the remote processing unit 17 computes in real time, through the computation module 23, the equivalent CO2 footprint of the machinery 3 using the formula:

[0031] It should be noted that such algorithmic transformation is linear, and that the variables Vi are not necessarily independent of one another.

[0032] In the event that the equivalent CO2 value exceeds a predetermined threshold, stored in memory means 21 of the remote processing unit 17, the latter will send to the edge device 7 a control signal 20 containing suitable commands intended for one or more actuators 5 of the machinery 3.

[0033] Such commands make it possible to control, whether directly or indirectly, the operation of the machinery 3, e.g. to adjust the speed of a motor, modify the air pressure, increase the steam temperature, adjust the solvent nebulization pressure, etc.

[0034] In this manner, the system 1 according to the invention offers the possibility of obtaining automatic and adaptive real-time feedback control impacting on the equivalent CO2 emission values associated with the machinery 3, and particularly on the processes executed by the latter and / or on the product obtained therefrom.

[0035] Optionally, the system 1 may send alert signals to an operator when such feedback control is performed.

[0036] Although the function that binds the equivalent CO2 footprint, CO2eq, of the machinery 3 to the plurality of variables Vi is known, such function cannot be fully implemented in a deterministic way because of the possible existence of complex correlations among the variables

[0037] One example of such complex correlations is the one existing between ambient temperature, ambient humidity, motor temperature and lubricant temperature.

[0038] The remote processing unit 17 comprises an artificial intelligence module 22 suitable for predicting the value of the dependent variable CO2eq as the independent variables Vi, change, and establishing which variables w of the machinery 3 should be adjusted and with how much intensity in order to bring the equivalent CO2 emission value back below a predetermined threshold value.

[0039] The artificial intelligence module 22 uses a supervised machine learning model, implemented by means of a regularized linear regression technique.

[0040] Regularization of the linear regression is necessary to counteract the overfitting phenomenon and mitigate the effects of multicollinearity, which may occur in that the presence of any mutual dependency among the values of the variables w is not known a priori.

[0041] Regularization is performed by adding a penalty that increases according to the complexity of the model. This means that a particularly complex model will be penalized more during the training phase. This ensures balance between the complexity of the model and its ability to generalize when it has to process new data for the first time.

[0042] In particular, the regularized linear regression technique employed is of the L2 type (also known as linear regression with Ridge regularization), wherein the regularization term is proportional to the sum of the squares of the model’s coefficients.

[0043] Such a model is particularly advantageous because it reduces the model’s sensitivity to the training data. This is due to the fact that said regression technique searches for coefficients that best approximate the data, thus making regression deviance as small as possible, which results in better generalization on the test data.

[0044] In other words, Ridge regularization is used to counteract the overfitting of an excessively complex model, which, during the training phase, would tend to store the training data instead of using them to learn, thus failing to generalize the data.

[0045] The training process involves the minimization of the sum of the squares of the errors between the values estimated by the model and the observed values of the dependent variable CCheq, along with the addition of the so-called “L2 penalty” to the cost function of the model, which is proportional to the sum of the squares of the coefficients Vi:

[0046] The parameter X performs the function of controlling the contraction of the penalty coefficients. For example, for = 0 the penalty has no effect, and the ridge estimates are equal to those of the ordinary least squares, whereas for k — > co the impact of the penalty grows, and the ridge estimates will converge toward zero.

[0047] Therefore, the choice of the parameter is of the utmost importance, and is made by means of cross-validation methods during the training phase, as will be explained below, using the technique called “K-fold cross-validation”.

[0048] Depending on the type of industrial machinery and application, the datasets used in the training phase may be either public datasets or datasets provided by machinery manufacturers. Some public datasets are, for example, those provided by associations and bodies supporting shared predictive maintenance (condition monitoring) projects, comprising: PHM Society Data Repository, UCI Machine learning Repository, NASA PHM Data Challenge.

[0049] Alternatively, the datasets may be provided directly by machinery manufacturers, who use them for maintenance applications.

[0050] Lastly, a third option is represented by the possibility of training the machine learning algorithm on the basis of real data directly measured on the machine. Although it may imply spending more time to create the dataset, this third option is the one that is most likely to lead to the most accurate results.

[0051] Once the reference dataset has been defined, the algorithm is trained on the basis of said dataset in order to iteratively “learn” the relationships between the input variables Vi and respective CO2 footprint values. This learning phase is carried out under “supervision”, i.e. in the presence of solutions (or labels) for the dataset used during the training, in the form of practical examples. The algorithm is thus trained on the basis of such examples, e.g. a matrix with labelled examples consisting of sets of pairs (x,y), where x represents the variables w and y represents their CO2 footprint. During the training, the algorithm learns to process a predictive model by estimation using a hypothesis function h(x) capable of approximating an unknown function f(x) that connects the input variables to the label thanks to the examples or solutions provided. By way of example, considering a scenario in which the operating temperature of the machinery 3 and the instantaneous electric power in KW drawn for its operation (expressed as the amperage read for each power phase multiplied by the respective voltage), the dataset that can be used for the training will comprise the data listed in the following table, where the last column contains the values of the equivalent CO2 footprint of said machinery 3 in three different working conditions:

[0052] Note also that operating temperature and drawn power should not be considered to be independent of each other, since in many kinds of machinery they are mutually bound by a positive, and sometimes non-linear, correlation of unknown formulation.

[0053] Once the hypothesis function h(x) has been determined, it is necessary to evaluate its accuracy by means of a test dataset, also provided under supervision.

[0054] For the present invention, the testing phase that allows evaluating the accuracy of the predictive model is carried out by “K-fold cross-validation” to minimize the previously described overfitting problem.

[0055] When “K-fold cross-validation” is used, a model is trained k times to select that value of which minimizes the prediction error over the k iterations.

[0056] Preferably, the value of k is chosen to obtain that each dataset training / testing group is large enough to be statistically representative of the larger dataset.

[0057] The value of k may be as high as n, i.e. the dataset size, which makes it possible to exclude just one sample from the dataset to allow each sample, during the iterations, to be used in the test dataset. However, it has been experimentally demonstrated that a value of k between 5 and 10, extremes included, provides a model estimate with low distortion and moderate variance.

[0058] The amount of data available for the training being equal, when values of k close to the previously specified lower extreme are chosen, the model is trained on smaller training datasets, which may lead to a more unstable model estimate (high variance). On the contrary, when values of k close to the upper extreme are chosen, the model is trained on larger training datasets, which may lead to better stability, but also to greater bias. In general, the larger the dataset used, the smaller the value of k can be.

[0059] When many computation resources are available, one may start by choosing a value of k=5 and submitting the first subsets to the model in order to verify if the performance evaluation metrics vary significantly. If result stability is achieved with this value of k, i.e. k=5, it will be sufficient to maintain such value to complete the training. Otherwise, it will be necessary to increase the value of k toward 10 until sufficiently stable results are obtained.

[0060] With reference to Figure 2, the following will describe a scheme of iterations (ITR) of subsets (Fold) of a dataset for training a supervised machine learning model used in the method according to the invention.

[0061] The phases envisaged by the k-fold cross-validation technique are:

[0062] (a) data preparation: the dataset, i.e. the set of all data, is divided into two partitions: one comprising the training data and one comprising the test data, as shown in Figure 2;

[0063] (b) data subdivision: the training dataset is randomly subdivided into k subsets (Fold) (with k=5 in the illustrative case of Figure 2), all having approximately the same size.

[0064] (c) iteration: the model is trained for a number of iterations ITR equal to the chosen value of k, wherein, for each iteration, the training dataset is in turn subdivided into an actual training partition for training the artificial intelligence module 22 and a performance validation (test) partition. The actual training partition for each iteration ITRx (where x represents the x-th iteration) is represented in Fig. 2 by the blocks “Fold i” (where i indicates the i-th block) in clear grey. The validation partition for each iteration ITRx is represented in Fig. 2 by the block “Fold i” in dark grey. At each iteration, therefore, one “Fold i” is used as the test dataset, and the other K-l “Fold i” are used as the training dataset. This means that the model is trained on the training data and evaluated on the test dataset. (d) evaluation: at each iteration, a different validation dataset is selected, the size of which is equal to the total size of the dataset / k, and the overall performance of the model is evaluated k times. In particular, the mean square error between the observed values and the expected values is computed. Lastly, once the best values for the parameters of the model to be adjusted (in the case under examination, the parameter to be used for L2 regularization) have been identified, the test partition is used for measuring the performance metrics over a portion of data that was not used during the training phase. The choice of the value of X to be used is made as follows: for each value of X considered, the mean of the prediction errors over the k iterations is computed. That value of X which minimizes the mean prediction error is then selected as the optimal value.

[0065] The training may be conducted automatically (via API) or with operator assistance, which can be provided, for example, through a user interface, e.g. available on a tablet aboard the machine, so that during the machine learning model training phase the operator can, through the user interface, select the data and submit them to the artificial intelligence system 22 in order to train it.

[0066] Feedback control over the variables w occurs directly via the edge device 7, through adjustment, by means of a feedback signal 10, of one or more actuators 5 connected to its output lines.

[0067] The advantages of the system and method according to the present invention are apparent from the above description.

[0068] The system according to the invention is advantageously capable of making an automatic computation aimed at adjusting one or more variables causing the predefined equivalent CO2 emission threshold to be exceeded. Thus, progressive, consequential, optimized and normalized actions can be taken on individual production variables on the basis of the feedback values measured on site in real time and of their impact on the quantity to be controlled, parameterized to CO2 emissions.

[0069] Of course, without prejudice to the principle of the present invention, the forms of embodiment and the implementation details may be extensively varied from those described and illustrated herein merely by way of non-limiting example, without however departing from the protection scope of the present invention as set out in the appended claims.

Claims

CLAIMS1. Method implemented through a computer (17), in particular a remote processing unit, for measuring and reducing the carbon footprint emitted by machinery (3), comprising the steps of:- receiving at least one variable Vi associated with at least one actuator (5) of said machinery (3) and relating to a phase of a manufacturing process;- reading, from a database (19), a plurality of computation parameters m for determining the carbon footprint of the phase of the manufacturing process;- computing the equivalent carbon footprint CCheqof the machinery (3) on the basis of the at least one variable w and said plurality of computation parameters m,- reading, from memory means (21), a predetermined threshold value;- when the value of the computed equivalent carbon footprint CCheqis higher than said predetermined threshold value:- providing, through a machine learning algorithm trained by means of a regularized linear regression technique, at least one command for the at least one variable w capable of reducing the value of the equivalent carbon footprint CCheq;- generating a control signal (20) comprising suitable commands to be applied to the at least one variable w for reducing the equivalent carbon footprint CCheq of said machinery (3).

2. Method according to claim 1, wherein said control signal (20) is supplied to an edge device (7), so that said edge device (7) will generate a feedback signal (10) applied to the machinery (3) and comprising the commands intended for one or more actuators (5) of said machinery (3) associated with the at least one variable w.

3. Method according to claim 1 or 2, wherein an alert signal is generated when said suitable commands are applied to the one or more actuators (5).

4. Method according to one or more of the preceding claims, wherein the at least one variable w comprises at least one of: temperature, pressure, speed, fluid level.

5. Method according to one or more of the preceding claims, wherein the plurality of computation parameters m comprises at least one parameter among: raw material type, raw material quantity, raw material origin, energy used during each manufacturing phase of the machinery (3), atmospheric emissions, water emissions, generated waste type, generated waste quantity, impact on human health, impact on the ecosystem.

6. System (1) for measuring and reducing the carbon footprint of machinery (3), saidsystem comprising an edge device (7) associable with the machinery (3) and connected, through connection means (16), to a computer (17), in particular a remote processing unit, wherein the edge device (7) comprises:- a plurality of sensors (9), coupled to a plurality of actuators (5) of the machinery (3), suitable for acquiring at least one variable w associated with at least one actuator of said plurality of actuators (5);- a microprocessor (13) configured for transmitting the at least one variable v / to the computer (17) and for receiving from the computer (17) a control signal (20) comprising the commands to be applied to the at least one variable w; and wherein the computer (17) comprises: a second microprocessor (18) suitable for receiving at least one variable w, generating a control signal (20) comprising suitable commands to be applied to the at least one variable w for reducing the equivalent carbon footprint of the machinery (3), said microprocessor (18) being in data communication with: o a database (19) suitable for storing a plurality of computation parameters otifor determining the carbon footprint of the phase of the manufacturing process of the machinery (3); o memory means (21) suitable for storing a predetermined threshold value; o an artificial intelligence module (22) suitable for storing a machine learning algorithm trained by means of a regularized linear regression technique and suitable for determining at least one variable w to be applied to one or more actuators (5) of the machinery (3) by means of suitable commands in order to reduce a value of the carbon footprint of machinery (3); o a computation module (23) suitable for determining an equivalent carbon footprint CCheq of the machinery (3) for a specific phase of the manufacturing process.

7. System (1) according to claim 6, wherein the microprocessor (13) is further configured for generating a feedback signal (10) applied to the machinery (3) and comprising the commands intended for one or more actuators (5) of said machinery (3) associated with the at least one variable w.

8. System (1) according to claim 6 or 7, wherein the edge device (7) further comprises a user interface, on which the microprocessor (13) displays the commands to be applied to one or more actuators (5) of the machinery (3) associated with the at least onevariable Vi.

9. System (1) according to one or more of claims 6 to 8, wherein said second microprocessor (18) is suitable for generating an alert signal when suitable commands are applied to the one or more actuators (5) of the machinery (3).

10. System according to one or more of claims 6 to 9, wherein said connection means(16) that connect the remote processing unit (17) to the edge device (7) comprise one of: a wireless connection, a LORA network, an Ethernet cable, an LTE / 5G network.

Images