FR3094504A1 - Digital device for regulating an industrial process with context - Google Patents

Abstract

Digital apparatus (1) for controlling an industrial process (20) comprising a plurality of machine elements (21) coupled to sensors (22) adapted to receive data (13) of operational parameters of the industrial process during the process. manufacture of products (30), the digital regulation apparatus (1) comprising a module (2) of technical specifications of products (30) to be manufactured by the industrial process (20), a module (3) of industrial parameters, receiving sensors (22) integrated into the industrial process (20), data (13) of operational parameters of the process during the manufacture of products (30) by the process, a module (4) for calculating the setpoint by automatic learning, and a module (5) for controlling machine elements (21). FIGURE 1

Description

Digital device for regulating an industrial process with context

The present invention relates to a digital apparatus for regulating an industrial process designed for the manufacture of a family of products, said industrial process comprising a plurality of machine elements designed to operate with ranges of values of setpoint parameters at the inside which the products of the family of products are likely to be manufactured, the machine elements being coupled to sensors adapted to receive data of operational parameters of the industrial process during the manufacture of the products.

Document EP3379357 describes an industrial process monitoring system using an advanced process controller (APC) connected to process status sensors and an industrial process anomaly detection module using models from machine learning (MLM) . The system makes it possible to generate alerts in the event of abnormal operation of the industrial process or the APC.

This type of system is based on one or more models attempting to represent the industrial process. Modeling is often a very complex process, making this approach tedious for complex industrial processes. There is also always a gap between the model and the actual process. Detecting errors in the operation of the process does not necessarily improve the process or the performance of the process.

The document EP0097053 describes a digital PID type regulation device intended to calculate a regulation signal in the form of an input signal of a process. This device makes it possible to carry out a regulation of a process during operation by detecting the error in relation to a setpoint. No means are provided to enable the process to be launched under optimal conditions or to optimize the process in progress. Furthermore, one or more instructions must be provided to the device.

There is therefore a need for a device which makes it possible to facilitate the start-up of a manufacturing process and to make it possible to optimize the process once it is underway.

To overcome these various drawbacks, the invention provides various technical means.

First of all, a first objective of the invention consists in providing a device making it possible to facilitate the launch of an industrial manufacturing process with a view to obtaining specific characteristics in the shortest possible time.

Another objective of the invention consists in providing a device which makes it possible to regulate an industrial process without having to use a modeling of this process.

Yet another object of the invention consists in providing a device which makes it possible to optimize the operational parameters of a manufacturing process.

To do this, the invention provides a digital device for regulating (or automatic piloting) an industrial process designed for the manufacture of a family of products, said industrial process comprising a plurality of machine elements designed to operate at the within ranges of target parameter values within which the products of the product family are likely to be manufactured, the machine elements being coupled to sensors adapted to receive data of actual operational parameters of the industrial process during the manufacture of products, said digital regulation device comprising:

i) a module of technical specifications of products to be manufactured by the industrial process;

ii) an industrial parameters module, receiving from said sensors integrated in the industrial process data of actual operational parameters of the process during the manufacture of products by the process;

iii) a setpoint calculation module by automatic learning, storing the operational parameter data of previous iterations of the industrial process in relation to the setpoint data applied to each machine element during these previous iterations, said setpoint calculation module by automatic learning being designed to define (without resorting to modeling of the industrial process) a set of setpoint values to be assigned to said machine elements according on the one hand to the technical specifications of a type of product to be manufactured and another part of all or part of the operational parameter data of previous iterations of the industrial process so as to optimize at least one of the operational parameters of the industrial process by using the operational parameter data of previous iterations of the industrial process for identical technical specifications or neighbors;

iv) a machine element control module, connected to the instruction calculation module by automatic learning, and to each of said machine elements, capable of controlling the industrial manufacturing process by supplying each of said machine elements with at least one setpoint value obtained by said setpoint calculation module by automatic learning.

Such an embodiment makes it possible to facilitate the launch of an industrial manufacturing process with a view to obtaining specific characteristics in the shortest possible time, without having to use a modeling of this process. Due to the imprecision of the models, and/or the great difficulty in modeling complex industrial processes often involving several thousand parameters, the fact of being able to rely on real historical data makes it possible to obtain reliable results, at advantageous costs.

According to an advantageous embodiment, the product technical specifications module comprises technical product specification data to be manufactured by the industrial process and operational parameter data to be optimized.

According to an advantageous variant, the industrial parameter module comprises non-adjustable parameter data coming from non-adjustable parameter data sensors of the industrial process and the set point calculation module comprises historical data of non-adjustable parameters.

The invention also provides a method for regulating (or automatic piloting) an industrial process designed for the manufacture of a family of products by a digital device for regulating (or automatic piloting) of said industrial process, said industrial process comprising a plurality machine elements designed to operate within ranges of setpoint parameter values within which the products of the product family are likely to be manufactured, the machine elements being coupled to sensors suitable for receiving operational parameter data of the industrial process during the manufacture of the products, said digital regulation apparatus comprising:

- a module of technical specifications of products to be manufactured by the industrial process;

- an industrial parameter module, receiving from said sensors integrated into the industrial process operational parameter data of the process during the manufacture of products by the process;

- a set point calculation module by automatic learning;

- a machine element control module, connected to the instruction calculation module by automatic learning, and to each of said machine elements, capable of controlling the industrial manufacturing process

said method comprising the steps according to which:

i) the automatic learning instruction calculation module receives technical specifications of a type of product to be manufactured;

ii) the set point calculation module by automatic learning receives one or more operational parameters of the industrial process to be optimized;

iii) the set point calculation module by automatic learning accesses the operational parameter data of previous iterations of the industrial process for technical specifications that are identical or similar to those previously received in step i;

iv) the set point calculation module by automatic learning calculates (without using modeling of the industrial process) a set of set point values to be assigned to said machine elements so as to optimize the operational parameter(s) of the industrial process previously received at step iii;

v) the setpoint calculation module by automatic learning transmits to said machine element control module the setpoint value(s) obtained;

vi) the control module pilots the industrial manufacturing process and provides each of said machine elements with the setpoint value or values which concern it;

vii) the machine elements carry out the manufacturing operations of the ordered product according to the instructions received.

According to an advantageous embodiment, the set point calculation module by automatic learning receives data from a plurality of non-adjustable parameters for the industrial process and determines the one having the greatest impact of deviation on the industrial process and searches for/calculates the most favorable operational parameters to correct or compensate or attenuate the deviation related to this non-adjustable parameter.

This embodiment makes it possible to set up operational modes taking into account the non-adjustable parameters (or contextual parameters), so that the industrial process undergoing these parameters is optimized.

By non-adjustable parameter is meant a parameter for which an instruction given to a machine element of the industrial process has no direct effect on this parameter. This can be due to the fact that the parameter cannot (or can hardly) be controlled by a machine element (weather, time parameter, etc.) It can be for example a physical value related to the environment in which the industrial process is implemented and on which no machine element can act, or a parameter extrinsic to the manufacturing process with its machine elements, such as for example the chemical composition or the origin of a raw material used in the manufacturing process. It can also be a parameter that can no longer be modified when the setpoint is calculated. For example, a reserve of paper pulp is manufactured at a certain moment, with a composition which is the result of adjustments, but once this reserve has been formed, the rest of the industrial process uses this reserve without modifying its composition. The adjustment parameters that were used for the production of the paste are then seen as non-adjustable parameters for the rest of the process.

According to an advantageous embodiment, the setpoint calculation module by automatic learning records and stores the operational parameter data of the iterations of the industrial process in relation to the setpoint data applied to each machine element during these iterations. This new data will then be used to further optimize the desired operational parameters.

Steps iii to vii are advantageously repeated one or more times until a stopping criterion is reached, such as for example the end of a production, obtaining a stabilized result, or other.

A latency time is advantageously provided between each new instruction. This latency time allows the technical effect linked to the new setpoint to have time to occur.

According to another embodiment, at least one operational parameter to be optimized corresponds to a technical specification of the product to be manufactured.

Advantageously, the optimization step is based on the analysis of the data of the operational parameters and the instructions of the previous iterations, to aim for a continuous improvement of the operational parameter to be optimized.

Also advantageously, the optimization or continuous improvement step comprises a sub-step of separating data from previous iterations of industrial parameters around a median.

The method advantageously comprises a maximization or minimization or stabilization step.

All the implementation details are given in the following description, supplemented by Figures 1 and 2, presented solely for the purpose of non-limiting examples, and in which:

Fig.1

is a schematic representation of an example of a digital control device connected to an industrial process;

Fig.2

is a functional flowchart of an example automatic control method.

FIG. 1 is a schematic representation of an industrial process 20 provided for the manufacture of a family of products 30. By family of products is meant a product which may have several variants or versions, these variants or versions being distinguished from each other. others by technical specifications of specific products.

The industrial process 20 is implemented by a plurality of machine elements 21. Each machine element 21 implements one or more manufacturing functions for which it is specifically designed. Each of the machine elements is controlled by one or more calculated set point values 16 . The values are generally included within ranges of setpoint parameter values within which the products of the product family are likely to be manufactured.

A module 5 for controlling machine elements 21 is connected to each of said machine elements 21 and is suitable for controlling the industrial manufacturing process by supplying each of these machine elements 21 with the calculated set point value(s) 16 making it possible to manufacture the desired product, with the desired technical specifications.

During the course of the industrial process, in particular during the manufacture of the products, the actual data 13 of the operational parameters making it possible to qualify and/or measure and/or monitor and/or supervise the process are captured or detected by a plurality of sensors 22 provided at the level of the machine elements 21 and/or in the environment of the industrial process.

The digital apparatus 1 for regulation (or automatic piloting) of the industrial process 20 makes it possible to automatically generate and supply the instruction(s) to the control module 5. It includes a module 2 of technical specifications of products 30 to be manufactured by the industrial process 20. This module groups together all the data 10 of technical specifications of the family of products that can be manufactured by the industrial process.

It also comprises a module 3 of industrial parameters, receiving the data 13 of effective or real operational parameters of the process during the manufacture of the products 30 by the industrial process. These data come from the sensors 22 integrated into the industrial process.

It also includes a set point calculation module 4 by automatic learning, described later. A microprocessor 6 and implementation instructions 7 allow the digital control device to operate autonomously.

In the example illustrated, this module stores the data 15 of operational parameters of previous iterations of the industrial process in relation to the data 14 of instructions applied to each element of machine 21 during the previous iterations of the industrial process.

The setpoint calculation module 4 by automatic learning is also designed to define a set of setpoint values 16 to be assigned to the machine elements 21 according to the technical specifications 10 of a type of product to be manufactured and the other part of all or part of the data 15 of operational parameters of previous iterations of the industrial process, so as to optimize at least one of the operational parameters 11 of the industrial process by using the data 15 of operational parameters of previous iterations of the industrial process for identical or similar technical specifications.

Examples 1 to 3, presented later, are used to describe and illustrate the optimization mode implemented by the setpoint module by automatic learning.

FIG. 2 illustrates the main steps of the industrial process regulation (or automatic piloting) method 20. At step 40, the set point calculation module by automatic learning receives the technical specifications of a product to be manufactured. At step 41, the same module receives the operational parameters of the industrial process to be optimized. It then accesses, at step 42, the operational parameter data of previous iterations.

In step 43, the setpoint calculation module by automatic learning calculates a set of setpoint values to be performed on the machine elements, and transmits to the machine element control module, in step 44, the setpoint(s) set values obtained. Using these setpoint data, in step 45, the control module controls the industrial manufacturing process and supplies the setpoint value(s) to the machine elements.

In step 46, the machine elements carry out the manufacturing operations of the ordered product.

According to one embodiment, the data 15 of operational parameters of the iterations of the industrial process are stored in relation to the data 14 of setpoints applied to each machine element 21 during these iterations. The fact of keeping the new data obtained, in addition to historical data, makes it possible to change the setpoint values with a view, for example, to continuous development. Thus, steps 42 to 46 previously described are repeated until a stopping criterion is reached, such as the end of production, or any other stopping criterion, such as obtaining a stabilized result, for example. It is also possible to provide a latency time after an instruction until the technical effect of the previous instruction occurs.

Various variants of the method are possible. For example, the operational parameter(s) to be optimized may correspond to one or more technical specifications of the product to be manufactured, such as for example the whiteness or the weight of a paper to be manufactured.

EXAMPLE 1 : application of a coat of paint

To illustrate the invention, a first simplified example, with a very small number of data, is presented. This first example concerns the manufacture of a toy, in particular a step of applying a coat of paint at a certain temperature. The part is therefore heated before or during the application of the paint, in order to improve the quality of the finish.

Parts transit at a speed v in an oven at temperature t in order to fix a paint on the part. The speed of the conveyor is imposed (depending on an upstream process for example) and only the temperature of the oven is adjustable. To have an optimal paint finish, the part must receive a certain amount of energy. If the speed is high, then it is necessary to increase the temperature value. Conversely, if the speed decreases, it is necessary to reduce the temperature value so that the total quantity of energy received by the part remains stable.

To illustrate this example, the amount of energy is equal to the temperature t divided by the velocity v. The amount of energy that maximizes the quality of the paint is t/v = 20.

In this example, the objective is to approach this optimal. We can illustrate such an objective as follows: Objective = 100 / (1+|20-(t/v)|).

Thus, for a maximum quality level of 100, which is the ideal to be achieved, we observe a deterioration in the quality (values below 100) of the painted parts as we move away from the optimal quality. .

By way of illustration, the history values only include four values, i.e. the last four parts or batches manufactured. The digital regulation device and the regulation process aim to define the setpoint values to be used for this production, or production phase without a preliminary model or device establishing the link between quality, temperature and speed.

The following data is the historical operational parameter data of the manufacturing line. Rows are sorted chronologically. We aim for a quantity of energy received per part as close as possible to 20.

part number Speed (m/s) Temp (°C) Qty Energy Skill level 1 5.2 80 15.4 17.8 2 5.7 85 14.9 16.4 3 4.1 90 22.0 33.9 4 4.7 85 18.1 34.3

Product no. 5 must now be painted, and the automatic regulation process makes it possible to provide a set point value according to the operational characteristic to be optimized. The four points in the history are separated into two populations around the median of the goal. Since we want to maximize the objective, the “good” pieces will be those of the group with a strong objective (pieces 3 and 4). Pieces 1 and 2 are therefore considered in this breakdown as being “bad” pieces. With this division, if we consider the temperatures, we obtain: 80° corresponds to a bad part, 85° corresponds to a bad and a good part, and 90° corresponds to a good part.

The interval that contains the most favorable data and the least unfavorable data or interval that maximizes the "crossrank" is therefore [85° 90°]. Its “rank” (or score) is (2/2-1/2)*100 = 50.

From this interval, the setpoint is extracted by taking for example the average: in the history, the following measurements correspond to the range of the rank: 85, 85, 90. The average (87.5°) is thus assigned as temperature adjustment setpoint for room 5.

part number Speed (m/s) Temp (°C) Qty Energy Skill level 1 5.2 80 15.4 17.8 2 5.7 85 14.9 16.4 3 4.1 90 22.0 33.9 4 4.7 85 18.1 34.3 5 4.1 87.5 21.3 42.7

As a variant, a value other than the average can be used, such as for example the median, the minimum or even the maximum depending on the case. A more complex operation can also be carried out on this interval to obtain an even more adjusted value, such as, for example, choosing the value of the interval for which the ratio of favorable data is maximum.

Since in this theoretical example, only the last four parts are taken into account, the first part is ignored, and the process continues with parts 2, 3, 4 and 5 which make it possible to obtain a new cut, and therefore another setting for room 6 (86.3°).

part number
1 Speed (m/s)
5.2 Temp (°C)
80 Qty Energy
15.4 Skill level
17.8 2 5.7 85 14.9 16.4 3 4.1 90 22.0 33.9 4 4.7 85 18.1 34.3 5 4.1 87.5 21.3 42.7 6 4.1 86.3 21.0 48.8

At each iteration, the best coins from recent history are used to generate the next setpoint to be applied based on these good coins. We observe that the level of quality progresses with each new instruction.

part number
1 Speed (m/s)
5.2 Temp (°C)
80 Qty Energy
15.4 Skill level
17.8 2 5.7 85 14.9 16.4 3 4.1 90 22.0 33.9 4 4.7 85 18.1 34.3 5 4.1 87.5 21.3 42.7 6 4.1 86.3 21.0 48.8 7 4.2 86.9 20.7 59.2 8 4.2 86.6 20.6 61.8

It is understood that this example is intentionally simplified for illustrative purposes. In addition, for this case, a simple modeling, makes it possible to identify the ideal temperature, t=20.v

So as soon as the speed is known, the ideal temperature can be derived. But such an approach requires modeling of the manufacturing process in progress. However, in practice, due to the great complexity of industrial manufacturing, such modeling is not always possible. When possible, it rarely gives an exact result because there are always disturbing factors (sometimes not measured) which make the modeling imprecise, or even false.

The digital regulation apparatus and the automatic regulation method of the invention make it possible to dispense with a model representing manufacturing. Disruptive factors, whether internal or external, are taken into account indirectly. For example, if there is wear on the insulation of the oven, the known method, which is based on modeling, does not take into account the quantity of energy lost, and therefore the level of quality of the painted parts does not is not optimal, because the energy lost due to the lack of insulation is not taken into account.

The process previously presented is based on a history including measurements where the insulation is already damaged, taking into account the impact on the level of quality. The calculated setpoints are therefore adjusted automatically.

EXAMPLE 2 : Whiteness of paper

The second example comes from the technical field of paper mills. As in any industry, what is manufactured (here paper) must meet specifications. Whiteness is part of the specifications that papermakers must meet, this example will illustrate the process of regulating this specification.

In a paper production line, the whiteness is measured by a camera which counts the dark spots present on the surface. A spot density is thus determined. The target value in this example is 100: the closer the value measured at the end of the process is to 100, the better.

These stains come partly from the composition of the dough. To limit the number of stains, it must be sufficiently well filtered. The operational parameter called “refusal ratio” makes it possible to act on this filtration.

Stains can also be caused by poor dough flow. An operational parameter is also associated with it.

Finally, the rate of recycled paper (%) used in the pulp is also an important parameter, but this parameter is suffered. It depends on external factors (economic, availability, etc.). No adjustment can therefore be made on this parameter, which has an impact. It is however recorded in the production history so that the settings of the other parameters (the operational ones) can be adjusted to take into account the variations of the parameters undergone. This process is called contextualization.

In this example the density of dark spots is modeled by the following function: reject ratio (%) / pulp flow rate (l/min) + rate of recycled paper / 10 - 2. This formula is given to follow the calculations of The example. In a real factory, as mentioned earlier, the spot density is measured, not calculated, and the underlying physical process is only partially known.

In the following table, a line corresponds to a roll of paper produced. This is for educational purposes, and any other production unit can be used instead.

The following table shows the initial data available for making paper of desired whiteness 100. Data for the last 4 reels is present.

reel number refusal ratio (%) flow rate (l/min) recycling ratio (%) stain rate 1 11.20 1,250.00 20.0 113.6 2 11.80 1,220.00 20.0 105.4 3 12.30 1,300.00 20.0 107.7 4 12.40 1,190.00 20.0 98.0

Coil no. 5 must now be produced, and the automatic regulation method makes it possible to provide a set point value according to the operational characteristic to be optimized. The four points of the history are separated into two populations around the median of the whiteness rate. The "good" reels will be the two with the stain rate closest to 100 (reels 2 and 4). Coils 1 and 3 are therefore considered in this breakdown as being “bad” coils. With this breakdown, if we consider the refusal ratios, we obtain: “11.20 corresponds to a bad reel, 11.80 corresponds to a good one, 12.30 corresponds to a bad one, and 12.40 to a good one.”

The non-adjustable recycling ratio parameter cannot be used to separate the good from the bad because it is constant. No contextualization is therefore carried out for the moment.

Considering only the refusal ratio, the interval that best separates the good reels from the bad reels is [11.80; 12.40]. In this example the module uses the average to assign a value. 3 reels are included in this range and their reject ratios are 11.80, 12.30 and 12.40. The average (12.17) is therefore applied to the operational parameter of refusal ratio.

A similar process is operated for the second operational parameter (throughput). The interval [1190, 1220] is therefore retained, and the average value 1205 is assigned to the operational parameter of the flow. These assignments are visible on line 5 of the following table.

All the operational parameters being affected, the factory can proceed with the manufacture of reel n°5 and the stain rate obtained is 99.

reel number refusal ratio (%) flow rate (l/min) recycling ratio (%) stain rate 1 11.20 1,250.00 20.0 111.6 2 11.80 1,220.00 20.0 103.4 3 12.30 1,300.00 20.0 105.7 4 12.40 1,190.00 20.0 96.0 5 12.17 1,205.00 20.0 99.0

In this example, a history of size 4 is used to carry out the piloting. Thus, to operate the settings of coil 6, coil 1 is ignored. Reel 5, on the other hand, is added to the history. Following the same process with the new history, reel 6 is produced according to the parameters shown in the table:

reel number refusal ratio (%) flow rate (l/min) recycling ratio (%) stain rate 1 11.20 1,250.00 20.0 111.6 2 11.80 1,220.00 20.0 103.4 3 12.30 1,300.00 20.0 105.7 4 12.40 1,190.00 20.0 96.0 5 12.17 1,205.00 20.0 99.0 6 11.99 1,212.50 20.0 101.1

The control continues in this way and gives rise to reels 7 and 8. To illustrate the contextualization process, a change in the rate of use of recycled paper occurs in reel 8. The following data is therefore obtained:

reel number refusal ratio (%) flow rate (l/min) recycling ratio (%) task rate 1 11.20 1,250.00 20.0 111.6 2 11.80 1,220.00 20.0 103.4 3 12.30 1,300.00 20.0 105.7 4 12.40 1,190.00 20.0 96.0 5 12.17 1,205.00 20.0 99.0 6 11.99 1,212.50 20.0 101.1 7 12.08 1,208.75 20.0 100.1 8 12.04 1,210.63 30.0 101.6

At each iteration, the stain rate approaches 100, which is the optimal production value. But the last reel, due to the high recycling rate, deviates from the production optimum.

Now reel 9 should be produced, still with a history of 4 reels. The 2 coils classified as good are 7 and 5.

Under these conditions, the impact of the non-adjustable parameters is now greater: coils produced with a recycling rate >=30 are all bad. Before continuing with the calculation of the operational parameters, a contextualization is therefore carried out: “recycling rate >= 30“ is added to the contextualization conditions.

The process of calculating operational parameters therefore starts again from the beginning, but with a different history: the history that corresponds to the new context. Thus, the last 4 reels that were produced with a recycling greater than or equal to 30% constitute the new history (the reel numbers no longer correspond: the reel formerly 8 corresponds to reel 4 now and the 1, 2, 3 come from past data and therefore were not visible until then).

With this new history closer to the current situation, the same process leads to the following output:

reel number refusal ratio (%) flow rate (l/min) recycling ratio (%) stain rate 1 13.00 1,250.00 35.0 97.7 2 12.00 1,220.00 30.0 102.7 3 12.30 1,200.00 35.0 99.1 4 12.04 1,210.63 30.0 101.6 5 12.17 1,205.00 35.0 100.5 6 12.24 1,202.50 35.0 99.7

The production therefore adapted to this change based on data closer to the present situation, and quickly modified the operational parameters to counter the change of a non-adjustable parameter in order to converge towards its production optimum: a stain rate close to 100.

EXAMPLE 3: Residual moisture content

In the field of chemistry, distillation columns are used to separate the constituent elements of a raw material. The raw material is subjected to a vapor which heats it and causes the gasification of the element sought. The process requires an input of heat which is delivered in the form of steam. Distillation columns are widely used in the chemical industry. From this process, a gas emerges which mainly contains the component sought by the industrialist. But it also contains impurities, including residual moisture.

The residual moisture content should be minimised, but it is not necessary to waste too much energy on this. From a certain threshold, the product is acceptable. In this example the threshold is set to 1.5%. If the final product contains more than 1.5% the quality will not be optimal, and if it contains less, energy will have been wasted and therefore the production cost will not be optimal (nor the environmental impact). A value around 1.5% is therefore sought.

In this example, the main operational parameter impacting the moisture content is the steam flow (T/h) which is used in the distillation column. In addition to the steam flow, another parameter also has an impact on the distillation process: the outside temperature. This parameter is of course not adjustable, but its measurement, recorded by sensors, is recorded in the history. To be able to illustrate the operation in a table, the following formula defines the impact of the airflow on the humidity level: humidity level = (airflow*4) - 18.5 - outside temperature / 50.

Like the following examples, a four-item history is used. In this case, it is a flow production, and therefore a line in the following table does not represent an actual element, but rather a production unit. A measurement is carried out at regular intervals, and the value of the measured sample feeds the history.

The history for this example is as follows:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.2 0 2.3 2 5.5 2 3.5 3 4.9 4 1.0 4 6.00 2 5.5

Before the measurement of sample 5 takes place, the regulation module has the possibility of acting on the plant by modifying the operational parameter of flow. For this, as in the previous examples, the history is used in order to obtain a more adequate setting. The goal is to approach a residual humidity rate of 1.5%.

The 4 points are separated into two groups: the two points closest to a rate of 1.5% together (they constitute the good points) and the others (the bad points).

Points 1 and 3 are therefore the two good points. The module deduces the interval which maximizes the separation of good from bad for the two parameters: flow [4.9; 5.2] and outdoor temperature [0; 0]. The flow interval is used to isolate two good samples. (samples 1 and 3 are good, and are in the flow rate interval [4.9; 5.2]) while the temperature interval allows only one sample to be isolated (only sample 1 is linked to a outdoor temperature of 0°). Since the flow parameter is an operational parameter, it will be used to carry out the adjustment. Here again, it is the average that will be used to extract the value of the interval. Only two samples are in the interval, and their mean is 5.05. The throughput operational parameter is therefore set to 5.05. Production will be impacted, and the resulting sample (#5) is shown in the following table:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.2 0 2.3 2 5.5 2 3.5 3 4.9 4 1.0 4 6.00 2 5.5 5 5.05 0 1.7

The first sample is then replaced by the last, then a new good/bad division is made to prepare sample 6. The good points are then numbers 3 and 5. The interval separating the best always relates to the parameter of bit rate, and it is [4.9; 5.05]. It contains only two points, and its average (4.98) is therefore assigned to the operational parameter of throughput. As the following table shows, this brings the residual rate even closer to 1.5:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.2 0 2.3 2 5.5 2 3.5 3 4.9 4 1.0 4 6.00 2 5.5 5 5.05 0 1.7

The process can then iterate using the new history. The following output is obtained:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.2 0 2.3 2 5.5 2 3.5 3 4.9 4 1.0 4 6.00 2 5.5 5 5.05 0 1.7 6 4.98 2 1.4 7 5.01 0 1.6 8 4.99 0 1.5

On this example again we can see how the process manages to bring production to a defined objective by acting on the operational parameters: here the residual rate of 1.5% is quickly obtained. And this without resorting to modeling. Only history processing is used.

Now, to illustrate the automatic contextualization mechanism, we assume that the manufacturing process of this type of product stops (to manufacture another product), then resumes a few months later when the external conditions have changed (see sample 9 to 20°). The output obtained becomes the following:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.2 0 2.3 2 5.5 2 3.5 3 4.9 4 1.0 4 6 2 5.5 5 5.05 0 1.7 6 4.98 2 1.4 7 5.01 0 1.6 8 4.99 0 1.5 9 4.99 20 1.1

As for the previous iterations, the last 4 samples are divided into two groups, the good ones being those with the residual rate closest to 1.5. Only the last sample (9) is therefore considered bad since it is the one with the most distant residual rate (6 and 7 are equidistant and 8 is perfect). Under these conditions, the best range for throughput is [4.98; 4.98], but this range contains only one of the 3 good samples. While the best range for the temperature parameter is [0; 2] and it contains all the good samples. The temperature parameter is therefore better at separating the good ones from the bad ones, so it is retained. But as it is a non-adjustable parameter, it is by definition impossible to act on it, the “temperature >= 20” context is therefore created. Then the process is restarted from the beginning, taking into account only the samples taken with outside temperatures above 20°, even if it means looking for older samples. The new history could then be this:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.46 25 2.8 2 5.48 20 3.0 3 4.75 27 0.0 4 4.99 20 1.1

With this history, the population is divided in two by taking as good, samples 4 and 1 which are closer to 1.5. And the ideal range found now relates to the flow rate [4.99; 5.46] which contains 2 good samples, and no bad ones. The average is therefore taken as an instruction: 5.23. The process iterates and will give, depending on the outside temperature, the following result:

sample number flow rate (T/h) outdoor temperature (°) residual rate (%) 1 5.46 25 2.8 2 5.48 20 3.0 3 4.75 27 0.0 4 4.99 20 1.1 5.23 22 2.0 5.11 21 1.5

In this example, the residual rate of 1.5% is reached in just a few iterations, even though the context has just changed. Now the “temperature >= 20” context is identified. It is therefore stored so that as soon as the outside weather changes, the context will also change, and the history used will correspond to the production weather.

1. Digital regulation device (or automatic pilot)
2. Product technical specification module
3. Industrial Parameters Module
4. Setpoint calculation module by automatic learning
5. Machine element control module
6. microprocessor
7. Instructions for implementation
8. --
9. --
10. Technical specifications of products to be manufactured
11. Operational parameters of the industrial process to be optimized
12. Parameter data not adjustable
13. Industrial process actual operational parameter data from sensors
14. Historical setpoint data applied to each machine element
15. Historical data of operational parameters
16. Setpoint value obtained by the setpoint calculation module
17. Historical data of non-adjustable parameters
18. --
19. --
20. industrial process
21. Machine element
22. Operational parameter sensor
23. Non-adjustable parameter sensor
24. --
25. --
26. --
27. --
28. --
29. --
30. Product to be produced

Claims (11)

Digital apparatus (1) for regulating an industrial process (20) designed for the manufacture of a family of products (30), said industrial process (20) comprising a plurality of machine elements (21) designed to operate at within ranges of target parameter values within which the products (30) of the family of products are likely to be manufactured, the machine elements (21) being coupled to sensors (22) suitable for receiving data (13) of actual operational parameters of the industrial process during the manufacture of the products (30), said digital regulation apparatus (1) comprising:
i) a module (2) of technical specifications of products (30) to be manufactured by the industrial process (20);
ii) a module (3) of industrial parameters, receiving from said sensors (22) integrated into the industrial process (20) data (13) of the effective operational parameters of the process during the manufacture of products (30) by the process;
iii) a set point calculation module (4) by automatic learning, storing the operational parameter data (15) of previous iterations of the industrial process in relation to the set point data (14) applied to each machine element (21) during these earlier iterations, said setpoint calculation module (4) by automatic learning being designed to define a set of setpoint values (16) to be assigned to said machine elements (21) according on the one hand to the technical specifications ( 10) of a type of product to be manufactured and on the other hand of all or part of the data (15) of operational parameters of previous iterations of the industrial process so as to optimize at least one of the operational parameters (11) of the industrial process in using the data (15) of operational parameters of previous iterations of the industrial process for identical or similar technical specifications;
iv) a module (5) for controlling machine elements (21), connected to the module (4) for calculating set points by automatic learning, and to each of said machine elements (21), able to control the industrial process of manufacturing by providing each of said machine elements (21) with at least one setpoint value (16) obtained by said module (4) for calculating setpoint by automatic learning. Digital regulation apparatus according to claim 1, in which the module (2) of technical specifications of products comprises data (10) of technical specifications of products (30) to be manufactured by the industrial process and data (11) of operational parameters to optimize. Digital control apparatus according to one of Claims 1 or 2, in which the industrial parameter module (3) comprises non-adjustable parameter data (12) originating from sensors (23) of non-adjustable parameter data of the industrial process ( 20) and the set point calculation module (4) includes historical data (17) of non-adjustable parameters. Method for regulating an industrial process (20) designed for the manufacture of a family of products (30) by a digital device (1) for regulating said industrial process (20), said industrial process (20) comprising a plurality of machine elements (21) designed to operate within target parameter value ranges within which the products (30) of the product family are likely to be manufactured, the machine elements (21 ) being coupled to sensors (22) adapted to receive data (13) of the actual operational parameters of the industrial process during the manufacture of the products (30), said digital regulation device (1) comprising:
- a module (2) of technical specifications of products (30) to be manufactured by the industrial process (20);
- a module (3) of industrial parameters, receiving from said sensors (22) integrated into the industrial process (20) data (13) of the effective operational parameters of the process during the manufacture of products (30) by the process;
- a module (4) for calculating the setpoint by automatic learning;
- a module (5) for controlling machine elements (21), connected to the module (4) for calculating instructions by automatic learning, and to each of said machine elements (21), capable of controlling the industrial manufacturing process ;
said method comprising the steps according to which:
i) the set point calculation module (4) by automatic learning receives technical specifications of a type of product to be manufactured;
ii) the set point calculation module (4) by automatic learning receives one or more operational parameters of the industrial process to be optimized;
iii) the setpoint calculation module (4) by automatic learning accesses the data (15) of operational parameters of previous iterations of the industrial process for technical specifications identical or similar to those previously received in step i;
iv) the setpoint calculation module (4) by automatic learning calculates a set of setpoint values (16) to be assigned to said machine elements (21) so as to optimize the operational parameter(s) of the industrial process previously received at the step iii;
v) the module (4) for calculating set point by automatic learning transmits to said one module (5) for controlling machine elements (21), the set point value(s) (16) obtained;
vi) the control module (5) controls the industrial manufacturing process and provides each of said machine elements (21) with the setpoint value or values which concern it;
vii) the machine elements (21) carry out the manufacturing operations of the product ordered according to the instructions received. Regulation method according to Claim 4, in which the module (4) for calculating set points by automatic learning receives data (12) from a plurality of non-adjustable parameters for the industrial process (20) and determines the one having the highest deviation impact on the industrial process and calculates the most favorable operational parameters to correct the deviation related to this non-adjustable parameter. Regulation method according to one of Claims 4 or 5, in which the set point calculation module (4) by automatic learning records and stores the data (15) of operational parameters of the iterations of the industrial process in relation to the data (14 ) instructions applied to each machine element (21) during these iterations. Control method according to claim 4, in which steps iii to vii are repeated until a stopping criterion is reached. Regulation method according to any one of Claims 4 or 5, in which a latency time is provided between each new setpoint. Regulation method according to any one of Claims 4 to 6, in which at least one operational parameter to be optimized corresponds to a technical specification of the product to be manufactured. Regulation method according to any one of Claims 4 to 6, in which the optimization step is based on the analysis of the data of the operational parameters and of the setpoints of the previous iterations, in order to aim for a continuous improvement of the operational parameter to be optimized . Control method according to claim 10, in which the optimization or continuous improvement step comprises a sub-step of separating data from previous iterations of industrial parameters around a median.