FR3140693A1 - Methods and systems for identifying workstation modifications - Google Patents

Abstract

The invention relates to a method of identifying modifications to be made to workstations. This method comprises the following operations: S10) transmitting to a computer a workstation record including different characteristics of the workstation to be improved; S20) on the basis of this recording, using a predictive model implemented by a computer, the modifications to be made to the workstation to be improved are identified; the predictive model being configured to indicate as an output the modification(s) to be made to the workstation based on modifications already made to existing workstations. Figure for abstract: Fig. 2

Description

Methods and systems for identifying workstation modifications

The present invention relates to methods and systems for identifying modifications to be made to workstations.

A workstation is a set of equipment generally including a seat, a work surface, most often one or more machines, as well as comfort equipment (lighting, heating, air conditioning equipment, etc.). A workstation is provided to accommodate at least one person to enable them to carry out a professional activity.

A workstation is generally designed to accommodate a single person to carry out the professional activity in question. However, in certain cases, a workstation can be provided to accommodate (simultaneously) two or more people to carry out this professional activity. For example, operators responsible for surface control operations on large parts (airplane wings, etc.) can share the same workstation.

The layout of the workstation naturally plays a vital role in the efficiency of the work carried out by the person(s) carrying out their professional activity at the workstation, as well as on the health of this person(s). ).

For this reason, there is a permanent need for methods and systems to improve workstations, in order to improve the quality of life at work for the professionals concerned, and to reduce as much as possible occupational illnesses and accidents likely to affect them.

To improve workstations, in a manner known per se, methods are used which define recommended arrangements for typical workstations.

However, these methods are relatively ineffective in determining desirable improvements for a large number of workstations, presenting distinct individual characteristics.

To remedy the limitations of existing methods of identifying modifications to be made to workstations to be improved, the following method is proposed.

This method is a method of identifying at least one modification to be made to a workstation to be improved, designed to accommodate at least one person to carry out a professional activity there. This method comprises the following operations: transmitting to a computer a workstation record comprising a plurality of characteristics of the workstation to be improved; on the basis of the recording of the workstation, using a predictive model implemented by the computer, at least one modification to be made to the workstation to be improved is identified; the predictive model being configured (for example, having been trained) so as to output said at least one modification to be made to the workstation based on modifications already made to existing workstations.

In the preceding definition, the term "computer" broadly designates any data processing device comprising at least one memory capable of recording data, and in particular a program, and one or more processors capable of executing this program. A computer can for example include one or more computers (personal computers, servers, etc.), located in the same place or distributed over several sites, etc.

By 'predictive model', we designate here a computer processing (that is to say a set of data processing operations) which allows, from initial data concerning a situation, to produce at the output information relating to this situation, but not yet available. A naturally predictive model is implemented using a computer.

In one embodiment, the predictive model is configured to output said at least one modification to be made to the workstation based on modifications already made to existing workstations, taking into account at least one individual characteristic of said at least one person carrying out their professional activity at the workstation (for example, height, visual acuity, etc.); and the identification of said at least one modification to be made to the workstation to be improved is carried out by taking into account at least one individual characteristic of said at least one person carrying out their professional activity at the workstation.

In certain implementation modes, modifications already made to existing workstations are modifications made to workstations similar to the workstation to be improved. A similar workstation is defined here as a workstation whose record has at least one characteristic having the same value, or a close value, or a value within the same range of values, as the record of the workstation of work to improve.

The characteristics taken into account to determine that two workstations are similar are preferably characteristics having a strong relationship with the level of risk at the workstation.

For this reason, preferably the characteristics taken into account to determine that two workstations are similar are risk variables, such as frequency of occurrence, severity, damage and detectability of a disaster likely to affect a person carrying out their professional activity at the work station in question.

In particular, said at least one modification to be made to the workstation can be identified by querying a database comprising workstation records relating to workstations similar to the workstation to be improved, the database indicating, for each similar workstation record, at least one modification possible or carried out for the workstation.

For the implementation of the method according to the present disclosure, different types of predictive models can be used.

The predictive model may include a model trained by machine learning to predict changes that have (actually) been made to existing workstations.

The predictive model may for example include a neural network. This neural network can for example be trained to predict at the output said at least one modification to be made to the workstation based on modifications already made to existing workstations, in particular similar workstations.

In certain embodiments, the predictive model comprises a model trained by machine learning to identify said at least one modification to be made to the workstation to be improved so as to reduce or even minimize a level of risk at the workstation to be improved.

In certain implementation modes, the predictive model includes a decision tree, and the identification of said at least one modification to be made to the workstation to be improved is carried out by browsing the decision tree.

The decision tree is preferably configured to allow the identification of modifications which minimize (or at least which reduce) a risk for the person(s) carrying out their professional activity at the workstation.

Any decision tree generation method can be used. Thus in certain implementation modes, the decision tree is a tree created by the CART algorithm or by the C.4.5 algorithm.

In certain implementation modes, the decision tree takes into account, as a final variable, a level of risk at workstations; and said at least one modification to be made to the workstation is selected so as to reduce or minimize a level of risk at the workstation to be improved.

The methods according to this disclosure are most often implemented in such a way as to reduce the level of risk at the workplace. The level of risk is a variable which characterizes the importance of the risk incurred by the person(s) carrying out their professional activity at the workstation. This level of risk is generally calculated as a function of several risk variables; these latter variables may include frequency of occurrence, severity, damage and detectability of a disaster affecting (that is to say, likely to affect) the person carrying out their professional activity at the workstation.

For this purpose, in certain modes of implementation, during the course of the decision tree, at least one choice is based on a risk variable chosen from a frequency of occurrence, a severity, a damage and a detectability of a disaster affecting the person carrying out their professional activity at the workplace.

In certain implementation modes, the predictive model is configured using a database comprising workstation records, said records comprising values of risk variables; a value of said at least one risk variable is recorded in the form of an element of a Fibonacci sequence, and/or is associated with a character string (ie, the values correspond to words or expressions (linguistic variables) such as 'never', 'rarely', 'often', 'always', etc.).

In some implementations, the database includes workstation records relating to workstations similar to the workstation to be improved, the database indicating, for each similar workstation record, at least a possible or carried out modification to the workstation.

In certain modes of implementation, the identification of said at least one modification to be made to the workstation to be improved comprises the following operations:

S20) using the predictive model, a list of workstations similar to the workstation to be improved is determined;

S40) said at least one modification to be made to the workstation is selected from said list of similar workstations.

In certain implementation modes, operation S40 includes the following operations:

S42) from the list of similar workstations, a list of possible modifications is determined;

S44) among the modifications in the list of possible modifications, said at least one modification to be made to the workstation is selected.

Step S44 can be carried out in different ways.

In certain implementation modes, during operation S44, among the modifications in the list of possible modifications, those which are most frequently used are selected: for example, only the N modifications which are used are selected (and retained). most frequently made (for example, the 3 most frequent modifications, or the 5 most frequent modifications, etc.) among the similar workstations in the list of similar workstations.

Step S44 can thus, for example, include the following operations:

- for each modification in the list of possible modifications, a number of times the modification has been made to a workstation in the list of similar workstations is calculated;

- said at least one modification to be made to the workstation is selected as being a modification or a set of modifications most or most frequently adopted for the workstation, possibly taking into account the professional activity and/or individual characteristic(s) of said at least one person carrying out their professional activity at the workstation.

• on détermine une baisse du niveau de risque pour le poste de travail, pour chacune des modifications de la liste de modifications envisageables ;
• on sélectionne ladite au moins une modification comme étant la ou les modification(s) qui permet(tent) la plus grande réduction du niveau de risque.

In other modes of implementation, in step S44:

• a reduction in the level of risk for the workstation is determined, for each of the modifications in the list of possible modifications;
• said at least one modification is selected as being the modification(s) which allow(s) the greatest reduction in the level of risk.

By extension, the present disclosure also includes a method of modifying a workstation. This method then includes a first step of implementing a method as defined above, so as to identify at least one modification to be made to the workstation considered; and a second step consisting of modifying this workstation by making said at least one modification identified in the first step.

The methods presented above are normally intended to identify and, if necessary, carry out modifications to be made to a set of workstations to be improved (for example, all the workstations on an industrial site, etc.).

The present disclosure also relates to a method for identifying modifications to be made to a set of workstations to be improved, in which modifications to be made to the workstations are identified by implementing an identification method as presented below. above, for each workstation; and, when implementing said method, the following operations are carried out:

for each of the workstations, (S20) using the predictive model, a list of workstations similar to the workstation considered is determined; determining said list of similar workstations comprising an operation (S26) in which a risk class is determined for the workstation considered, from the risk levels of the workstations of said list of similar workstations; And

S30) an order of processing priority is determined for the workstations, based on the risk classes determined; Then,

the operations (S40) of identifying said at least one modification to be made, for the workstations, are carried out according to the order of priority determined in operation S30.

A risk class is defined here as an estimate of the level of risk, for the workstation considered.

In the method presented above, determining an order of priority makes it possible to determine and carry out the modifications to be carried out as a priority for the workstations with the highest risk class.

In a particular embodiment, the different operations (or steps) of the methods presented previously are determined by computer program instructions.

Consequently, the present disclosure also relates to a computer program comprising instructions which, when the program is executed by at least one processor, cause said at least one processor to execute the operations of a method presented previously. This program may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable shape.

The present disclosure also relates to a non-volatile computer-readable storage medium, on which the computer program presented above is stored. The information carrier can be any entity or device capable of storing the program. For example, the support may comprise a storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or even a magnetic recording means, for example a floppy disk or a disk. hard. Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in executing the method in question.

By extension, the present disclosure also relates to an identification device comprising at least one processor, as well as a storage medium as defined above.

There schematically represents a system for identifying modifications to be made to workstations according to the present disclosure;

There is a flowchart representing the first steps of a first method of identifying modifications to be made to workstations according to the present disclosure;

There is a flowchart representing the following steps of the first method of identifying modifications to be made to workstations according to the present disclosure;

There is a flowchart representing the steps of a second method of identifying modifications to be made to workstations according to the present disclosure; And

There is a schematic representation of a decision tree used in the first method of identifying changes.

Description of embodiments

By way of example, a system and methods for identifying modifications to be made to workstations according to the present disclosure will now be presented in relation to Figures 1 to 4.

Although the examples presented relate to a workstation intended for a single person, it is understood that the systems and methods according to the present disclosure can also be used for workstations intended for several people.

There thus represents a system 10 for identifying modifications to be made to workstations. The system 10 presents the hardware architecture of a computer, as illustrated schematically on the . Any computer may be used to implement the methods presented in this disclosure.

The system (or computer) 10 notably comprises a processor 12 and a non-volatile memory 14.

The non-volatile memory 14 of the system 10 constitutes a recording medium conforming to the present disclosure, readable by the processor 12 and on which a computer program P conforming to the present disclosure is recorded. In accordance with the present disclosure, the program P includes instructions for executing the steps of a method for identifying modifications to be made to workstations and more broadly, includes instructions for executing the steps of a method of identifying modifications to be made to a set of workstations.

The methods and systems according to the present disclosure allow the monitoring and modification of a workstation, or a set of workstations, so as to improve these workstations, and in particular in a manner most often to reduce the level of risk to which people carrying out their professional activity at these workstations are exposed.

The implementation of these methods and systems makes it possible to evaluate and progressively reduce the level of risk of all the workstations considered, iteratively, as follows.

A first and a second method according to the present disclosure, using predictive models of different types, will now be presented. In the first method, the predictive model is a decision tree, which makes it possible to identify the preferable modifications to be made to the workstations. In the second method, the predictive model is a neural network, which is used to perform the same function.

First method: determination of modifications by decision tree

A. Preliminary step: creating a BD database

Recording of workstations, people, professional activities

The first step, for implementing the methods according to the present disclosure, generally consists of establishing a BD database.

In the embodiment presented, this database is a relational database which associates records relating to workstations. In the embodiment presented, this database also includes records relating to people and records relating to professional activities carried out.

A record for a workstation, also called a workstation record, is data that brings together different characteristics of the workstation in question. It can include, for example, the location of the workstation (address, building, floor, room number, company, department, etc.). It generally indicates a category of which the workstation belongs, chosen from a predefined typology of workstations (industrial, administrative, etc.).

Workstation characteristics mainly include characteristics relating to the layout of the workstation.

These characteristics may include, for example, characteristics of the furniture making up the workstation; geometric characteristics of the workstation (geometry of the desk or work surface, and/or the seat); characteristics relating to lighting (position, power, characteristics of the light emitted, for each light source); etc.) ; characteristics of a machine used at the workstation (type (machining center), dimensions, noise characteristics, electrical characteristics, etc.); etc.

Workstation characteristics may include:

- information relating to a type of environment (office/factory; individual office/open space);

- the average temperature ; the sound level; the type of floor (tiles/concrete/..); the type of electrical installation (bare cable / under trunking / recessed; the professional activity carried out (tasks performed; postures during work (and time spent in each posture)).

The record for a workstation preferably also indicates, if this information is available, the modification(s) previously made to the workstation (for example: a height-adjustable workstation desk was installed to replace the workstation desk of fixed height, on such date).

A modification is a material change made or to be made to the workstation, normally to improve it.

A modification, for example, may consist of:

- add motorization to adapt the workstation to the person's body shape.

- add ventilation/air extraction.

– modify the layout (height or depth of the work surface, distance between the person and the screen, orientation of the screens, etc.); choice of screen…; etc.

A registration for a person may in particular indicate health data of the person in question, when such data is available.

A registration for a professional activity carried out may include a list of tasks usually carried out within the framework of the professional activity in question (e.g. computer work, welding, etc.).

The BD database can potentially include hundreds of thousands or millions of workstation and person records.

Recording of risk data

From the records recorded in the BD database, statistical data is calculated, also recorded in the BD database.

First of all, for each workstation, we identify as completely as possible all the risks to which the person(s) working at the workstation are exposed: exposure to screens, exposure to asbestos, exposure to high temperatures, vibrations, etc.

We therefore establish for each workstation the list of risks – or rather, exposures Em (m=1… M) to risks, indicating all the identified risks to which the person(s) carrying out their professional activity at the workstation is or are exposed. of work considered. This list is saved in the record of the workstation considered. By ‘exposure to a risk’ we mean exposure to a certain type of risk at a workstation in question.

Exposure is normally quantified. It can also be standardized, so as to have the same value for all people having a professional activity at a workstation of a type considered (independently of the individual characteristics of the person).

• Le type de métier ou le métier exercé par une personne exerçant une activité professionnelle au poste de travail considéré ;
• la catégorie de la personne exerçant une activité professionnelle à un poste de travail du type considéré, une catégorie de personnes étant définie par une ou plusieurs caractéristiques personnelles spécifiques (sexe, âge, pathologie,..) communes à la catégorie de personnes.

However, in certain embodiments the exposure is determined more precisely, taking into account additional criteria, such as in particular:

• The type of profession or occupation carried out by a person carrying out a professional activity at the work station in question;
• the category of the person carrying out a professional activity at a workstation of the type considered, a category of people being defined by one or more specific personal characteristics (sex, age, pathology, etc.) common to the category of people.

Then, for each PTi workstation, the values of one or more risk variables are determined and recorded in the BD database, for each Em exposure identified for the workstation considered.

The following risk variables can be determined: The frequency of occurrence F, the severity G, the damage DO (The damage can be crushing, death, disability, etc.) of a claim for the risk considered. DE detectability (variable representing the probability of detection of a disaster that has occurred, per unit of time) can also be used.

The values of risk variables can be determined in different ways, in order of decreasing preference (or precision):

a) they can be determined directly from available historical data relating to losses occurring during the past period;

b) they can be calculated from other data in the database (for example by analogy) (for example, the frequency of exposure to a risk can be calculated from all workstations belonging to a same type of workstations, or having certain characteristics in common); Or

c) they can be determined from pre-established tables.

The second option (b) can be chosen if the database has sufficient data to establish a reliable value for a risk variable: this value can then be calculated from the database data. The value can for example be calculated by taking into account all the people carrying out a professional activity on a workstation of the type considered, for all the professional activities carried out on the workstation.

The third option (c) can be used especially when options (a) and (b) are not available. The value of the risk variable is then determined from pre-established tables indicating recommended values for certain risk variables, for the risk exposures considered. This third option then generally uses a table of business codes. Such a table of occupational codes is a relational table indicating, for a certain type of workstation, for a profession considered, one or more risks to which a person exercising the profession considered at the workstation considered is exposed, or may be exposed. . The business table possibly includes, for the risks it mentions, the values of the risk variables for these different risks.

In the BD database, in certain embodiments, the values of certain variables, in particular of certain risk variables, can be recorded in the form of discrete (non-continuous) variables.

In this preferably case, if the risk variable is a numeric variable (i.e., a variable that normally is or can be represented by a number), at the time the value of the variable must be entered or selected by the operator responsible for constituting the database, it can be provided that only certain values of the variable considered are offered to the operator; in particular, the operator can be offered values which are not separated two by two by constant gaps, and are even preferably separated by increasing gaps, in particular defined using Fibonacci sequences.

This entry mode facilitates the choice of appropriate values while only offering a small number of values to the operator, which makes it easier for him to choose the most appropriate value.

For the same reason, for entering a numerical variable, it can be provided that the value of the variable is selected by the operator by a choice between several expressions (words or groups of words), rather than between several values digital. This situation occurs in particular when certain values that we wish to record in the database are not measured by sensors and must therefore be entered by the operator.

For example, the vibration level at a workstation (as an example of a numerical variable) may not have been measured. In this case, it may be necessary or at least useful that a value of this vibration level is nevertheless entered into the database.

For this purpose, the operator (who is in this case preferably a person working at the workstation, and therefore who is exposed to vibrations) can be asked to choose from several expressions (word or group of words) the one which best corresponds to the vibration level at the workstation in question. For example, the operator may be asked to choose from “very weak” “weak” “medium” “strong” “very strong” the expression that best corresponds to the vibration level at the workstation in question.

The chosen expression can then optionally be converted into a numerical value.

When for each PTi workstation, the values of the risk variables have been recorded for the different exposures Em identified for the workstation considered, the risk level NR(PTi) at the PTi workstation considered is calculated.

For each PTi workstation and taking into account the different exposures to Em risks identified for the type of workstation, the risk level NR(PTi) at the workstation is determined and recorded in the BD database. PTi considered.

This level of risk NR(PTi) is a variable which quantifies, using a single value, the overall level of risk incurred by a person (or one of the people) carrying out their professional activity at the workstation in question.

Different formulas or functions can be used to determine the level of risk based on the value(s) of the risk variable(s). In general, the risk level is obtained by one of the following formulas:

NR(PTi) = F x G x DO or

NR(PTi) = F x G x DO x DE if the detectability of disasters is taken into account.

Finally, the overall risk level for all workstations is calculated by taking the arithmetic average of the risk levels NR(PTi) of all the workstations considered.

Recording of desirable modifications to workstations and/or modifications made to them

In addition, for each PTi workstation, when possible, we collect and record a list of desirable modifications for the workstation, and/or the list of modifications that have been made to the workstation (preferably with their date of completion). By ‘desirable modification’, we generally mean a modification which reduces the level of risk for the workstation in question.

In certain embodiments the default database can be configured so as to make it possible to determine one or more desirable modifications for each workstation, or at least for a large part of the workstations registered in the base.

For this purpose, namely to make it possible to determine one or more desirable modifications for a workstation, the database can for example include typical values (average values and/or minimum or maximum limit values, considered desirable) for different characteristics of workstation records; particularly for recordings of typical workstations.

For example, the database may include a typical record for an office workstation, at which the person works on a computer, that is to say a record serving as a reference to follow, for office workstations.

In this case, for the typical recording for an office workstation, the database may include typical values for different characteristics, such as for example a minimum desirable height for the work surface, minimum and maximum desirable distances between the head of the person occupying the workstation and the screen, etc.

In this case, for a workstation record, one or more desirable modifications can be determined as the change(s) to be made to the workstation to ensure that its characteristics match the values of the record for the typical workstation.

The BD database can be structured so as to indicate, for a given risk exposure, one or more modifications that can be applied to the workstation considered so as to reduce the exposure to a risk considered. For example, for exposure to a risk of loss of visual acuity due to working on a screen, the database can associate with this exposure the modification consisting of increasing the size of the screen, the modification consisting of using a screen and/or a seat whose position is adjustable, etc.

The BD database can also be structured in such a way as to indicate, for a risk considered, a list of accident(s) or occupational disease(s) associated with the risk considered.

In certain embodiments, for at least part of the desirable modifications, the desirable modification is recorded in relation to one or more exposures to a risk identified for the workstation considered, possibly with the importance of the reduction of the risk level NR(PTi), and/or the reduction in the risk of accident or occupational disease, caused by the modification. This latest information makes it possible to calculate the reduction in the rate of accidents or occupational illnesses that can be achieved by making certain modifications to the workstations considered.

Once the BD database has been created, all of the workstations taken into account in the work base can be monitored; modifications to be made to these workstations can be determined by following the methods proposed by this disclosure in order to reduce the overall level of risk for all workstations.

In the first method, the creation of a database (as presented above) constitutes a first preliminary step to be carried out to allow the identification of the necessary modifications to the workstations. In the implementation method presented here, a second preliminary step is necessary: the establishment of a decision tree.

B. Second preliminary step: establishing a decision tree

Once the database has been established, that is to say when the statistical variables and the risk variables have been calculated and the desirable and/or carried out modifications recorded, a decision tree is established from the database. .

The input data to this decision tree is the set of records for PTi workstations, or possibly a subset of this set. The output variable or ‘critical variable’ of the decision tree is the risk level NR(PTi) at the workstation.

Each of the variables in the workstation records can be used as a determining parameter for traversing the decision tree (decision tree traversal parameter): for example, the 'workstation vibration level' (for a station on industrial machine), etc.

The parameters used to browse the decision tree are the variables which will be used, as will be explained later, to identify workstations similar to the workstation considered.

For this reason, the parameters most influencing the level of risk at the workstation are chosen as a priority as parameters for traversing the decision tree.

Preferably, at least one risk variable is used as a parameter for traversing the decision tree.

During the construction of the decision tree, it can be determined that certain variables that are part of the workstation registration do not influence (or very little) the choice of the recommended modification(s). In this case, the decision tree is reconstructed without taking these variables into account.

The construction of the decision tree is carried out using an iterative algorithm. Any known decision tree construction algorithm can be used, such as CART, C4.5, etc.

Following such an algorithm, most often at each iteration, the data (i.e. PTi workstation records) are divided into several groups.

The first division is obtained by choosing the explanatory variable which allows these records to be better separated. This division gives subclasses (or subgroups) attached to the root of the tree.

The division process is then repeated several times for each subclass that corresponds to one of the previously calculated subgroups, until the separation process stops.

The output of the algorithm produces a tree that establishes hierarchical relationships between variables.

The decision tree thus formed therefore makes it possible to automatically identify workstations similar to the workstation considered. Thus, when a record for a PTi workstation to be improved is provided as input to the decision tree, the latter provides as output a list of workstations ({Ptk}, k=1…K) similar to the PTi workstation to be improved.

C. Identifying and carrying out modifications for a set of workstations

The decision tree obtained then makes it possible to implement a method according to the present disclosure to identify, then to carry out, the preferable modifications for a set of workstations considered to be improved (This set of workstations may include workstations database work, and used to generate the decision tree, and/or new workstations not recorded in the database).

For each of the workstations of the set of workstations to be improved, initially the steps S10 and S20 described below are carried out ( ). This makes it possible to identify, for each of these workstations, an LPS list of similar workstations, and optionally a risk class for the position considered.

S10) Entering input data

We enter the data concerning the workstation to be improved (PTi), constituting a record of the workstation, and optionally we enter the data concerning the person(s) carrying out their professional activity(ies) at the workstation. to improve, constituting a person record.

S20) Determination of an LPS list of similar workstations

In step S20), the following operations are carried out:

In a first step S22), from the database BD, an LPS list of workstations similar to the workstation to be improved is generated.

This LPS list includes workstations that have similar characteristics to the workstation to be improved.

In this mode of implementation, to determine this LPS list of similar workstations, we go through the decision tree. The latter is provided as input with the data from the recording of the PTi workstation as well as the data of the person or persons carrying out their professional activity at the PTi workstation.

We then traverse the decision tree, so as to select at each node the branch including the workstation records having, for the determining parameter evaluated at the node considered, the same value as the PTi workstation record that l we want to improve.

As an output, the decision tree therefore provides an LPS list of workstations similar to the PTi workstation.

An example of traversing the AD decision tree represented by the is detailed (in part) below.

We provide a workstation record as input to the AD tree shown on the .

At the first node of the tree (node N1, or root), it is determined whether for the workstation record considered, the frequency f1 (of occurrence of a given type of disaster, for example a fall disaster ) is less than or equal to 2 (2 is an arbitrary value, used here as a reference value as an example), in which case the algorithm continues at node N2, or greater than 2, in which case the algorithm continues at node N21. In this example, the algorithm continues at node N2.

At node N2, it is determined whether for the workstation record considered provided as input, the severity g1 of the claims (for a type of claim considered at node N2, in this example, a fall) is less than or equal to 2 , in which case the algorithm continues at node N3, or greater than 2, in which case the algorithm continues at node N10. In this example, the algorithm continues at node N3.

At node N3, it is determined if for the workstation record considered provided as input, the frequency f2 (of occurrence of a second type of disaster, for example an electrocution) is less than or equal to 1, in which case the algorithm continues at node N4, or greater than 1, in which case the algorithm continues at node N7. In this example, the algorithm continues at node N4.

At node N4, it is determined whether for the workstation record considered provided as input, the severity g2 of the claims (for the second type of claim considered at node N3, in this example, the risk of electrocution) is lower or equal to 1, in which case the algorithm continues at node N5, or greater than 1, in which case the algorithm continues at node N6. In this example, the algorithm continues at node N5.

Node N5 is a terminal node. When the algorithm arrives at a leaf node, it outputs a list of workstations associated with the leaf node; by construction of the decision tree, this list is the LPS list of workstations similar to the workstation considered provided as input to the decision tree AD. In the example presented, at node N5 the LPS list of similar positions returned by the AD decision tree includes a number n (n = 93) of workstation records.

S24) Determination of the number of similar workstations in list L

In a second step S24), the number n of workstations similar to the PTi workstation considered is determined: this is the number of elements n of the LPS list returned by the decision tree (n=93 in the example present).

It is then evaluated whether the number n of similar positions falls within a range of values which makes it possible to effectively identify improvements for the workstation.

S241) Case where the number n of similar positions identified is too low

In some cases, the decision tree outputs a number n of similar workstations which is too low; for example, no similar positions are identified. In this case, we relax the similarity criterion, for example by reducing the list of characteristics defining the workstation provided as input to the decision tree. For example, we can delete characteristics that do not seem to have an impact on the level of risk at the workstation considered.

Reducing the list of characteristics defining the workstation therefore leads to obtaining a simplified workstation record PTi_simpl.

The PTi_simpl record of the workstation simplified by deleting certain data is then provided as input to the decision tree.

The simplification carried out allows the path of the decision tree to be interrupted at a lower depth than with the initial PTi recording for the workstation; consequently, the _simple number n of similar workstations identified in the LPS list by the decision tree on the basis of the registration of the workstation thus completed is generally higher than the number n of similar workstations determined initially.

S242) Case where the number n of similar positions identified is too high

Conversely, in certain cases, the output decision tree identifies too many similar workstations n. In such cases, we therefore want to reduce the list of similar workstations. For this purpose, additional information concerning the workstation is preferably collected. This information may be requested from the person(s) occupying the workstation; or we can use additional data sources: occupational physician, ergonomist, etc.

The PTi_compl record of the workstation completed using this additional information is then provided as input to the decision tree.

This additional information makes it possible to browse the decision tree to a greater depth; consequently, the _complete number n of similar workstations identified in the LPS list by the decision tree on the basis of the recording of the workstation thus completed is normally lower than the number n of similar workstations determined initially.

When the number n of similar workstations is within the desired value range, the algorithm continues with step S26.

S26) Determination of a risk class for the workstation considered

When the number of similar positions identified is considered acceptable, operation S26 is carried out.

During operation S26, from the LPS list of similar workstations, obtained at the terminal node reached, the value y of the risk class is calculated. This risk class value is a variable representative of the average risk level of the workstations in the list of records obtained for a terminal node. It can be calculated, for example, by averaging the risk level of the workstations in the list of similar workstations.

In the example presented, the value obtained for the terminal node N5 (y=1) means a very low average risk level for the recordings of workstations similar to the workstation considered.

Once operations S10 and 20 have been carried out for each of the workstations, the algorithm continues with operation S30 ( ).

S30) Determination of the order of priority for workstations

During operation S30, an order of processing priority is determined for all of the workstations, based on the risk classes determined for each of them.

S40) Determination of preferable modifications for workstations

We then carry out for each of the workstations of all the workstations (or possibly only for part of them), an operation (S40) during which we identify the modification(s) (preferable) to bring to workstations.

The order in which operations S40 are carried out depends on the order of priority determined in operation S30. For example, operations S40 are carried out in the order of priority calculated in step S30.

Operation S40 indicated above consists of selecting one or more modifications to be made to a workstation, from the LPS list of workstations similar to this one established during operation S20.

Operation S40 can be carried out by successively carrying out the following two operations S42 and S44:

Operation S42: From the LPS list of similar workstations, we first establish an LME list of possible modifications for the PTi workstation. This LME list can for example include all the modifications made to at least one of the workstations in the LPS list.

A possible modification for the workstation may also be a modification recorded in the BD database as a possible modification for a typical workstation, corresponding to the workstation considered.

For example, the database may include as a typical workstation, an office type workstation. In this case, the possible modifications recorded may be: changing the height of the screen, changing the arrangement of the seat and the work surface so as to change the person's working position, etc.

The LME list may possibly include, for each possible modification:

- a proportion of the number of times this modification has been adopted for workstations on the list (i.e.: number of cases of implementation of the modification considered / number of workstations on the list LPS); and or

- a reduction in one or more exposure(s) to a risk caused by the modification considered.

Operation S44: among the modifications of the LME list of possible modifications, one or more modifications considered preferable are selected.

In some cases, the number of possible modifications listed in the LME list is quite small. In this case, we can possibly directly determine the modifications to be made, without a prior optimization phase.

However, most of the time, the number of possible modifications listed in the LME list is relatively large. In this case, preferable modifications can be selected in two ways.

Following a first approach, the modifications selected as preferable are the modifications which appear most often among the possible modifications.

For example in this case, step S44 may include the following actions:

- for each modification of the LME list of possible modifications, a number of times the modification has been made to a workstation in the list (LPS) of similar workstations is calculated;

- said at least one modification is selected as being a modification or a set of modifications most or most frequently adopted for the workstation, possibly taking into account the job performed and/or individual characteristic(s) s) of the person(s) stationed at the workstation.

Following a second approach, the modifications selected as preferable are identified in the following way.

In this case, the database BD includes, for each workstation record, a risk level NR(PTi) at said workstation PTi or data making it possible to determine a level of risk (NR(PTi)) at said workstation of work (PTi). In step S40 (and in particular when it takes place, as in the example presented, in step S44), the preferable modification(s) are selected so as to reduce or minimize the level of risk (NR( PTi)) at the workstation to be improved.

Concretely, for each modification Mj possible for the workstation PTi, we calculate the reduction in the risk level ΔNRij obtained when the modification Mj considered is carried out at the workstation to be improved PTi in the following manner.

We first determine the current risk level NR(PTi) for the workstation to be improved PTi.

We then determine the risk level NR(PTij) of the workstation PTij which is the workstation PTi modified by making the modification Mj envisaged.

The risk level NR(PTij) can be determined for example by providing as input to the decision tree the record PTij of the workstation PTi after carrying out the modification Mj: the decision tree indicates as output the list of PTk workstations similar to the modified PTij workstation. We then calculate the risk level NR(PTij) for the modified workstation as being the average risk level of the workstations PTk similar to the modified workstation PTij.

We then calculate the reduction ΔNRij in the risk level for the position PTi as being equal to the difference: ΔNRij = NR(PTij) – NR(PTi).

We then determine an LMP list of preferable modifications, which are the modifications considered to be made as a priority to the workstation in question.

This LMP list of preferable modifications can for example be established as the list of possible modifications for the workstation PTi for which the reduction ΔNRij in the level of risk for the workstation PTi is the greatest (for example, is greater than a threshold , the threshold can be predetermined or determined iteratively as being the lowest threshold value for which the number of preferable modifications retained is greater than a certain predetermined minimum number).

S50) Carrying out modifications

The workstation PTi is then modified by carrying out one or more of the preferable modifications identified in step S40. By making this or these modifications, the workstation is generally improved. In particular, if the chosen improvements were chosen so as to reduce the level of risk, the realization of the chosen improvement(s) at the workstation reduces the level of risk at the PTi workstation.

Updating information

The modifications made at the PTi workstation are then recorded in the BD database.

The algorithm then resumes at steps A and B: The global database BD is updated, and the decision tree AD is updated.

The method proposed above allowing the choice of modifications to be made to workstations and the carrying out of these modifications is preferably implemented repeatedly, for example at fixed intervals.

Second mode of implementation: direct determination of modifications using a pre-established digital model

In the first mode of implementation presented above, the predictive model mainly comprises a decision tree obtained using a BD database, comprising records of existing workstations, indicating modifications already made to them.

However, it has become apparent that models trained by machine learning, using a learning base comprising modifications already made to existing workstations, can also exhibit remarkable performance.

In the second mode of implementation, which will now be presented, a predictive model is configured (trained) to directly establish the list of the preferable modification(s) to be made to a PTi workstation to be improved. This list is established directly from the recording of the workstation considered, that is to say without it being necessary to first identify a list of workstations similar to the workstation considered.

The procedure to follow to identify and make changes for a set of workstations is as follows:

Preliminary step: Establishment of the predictive model

A predictive model configured to output a list of preferable modifications to be made to a PTi workstation to be improved, based on the registration of this workstation, is established.

This predictive model can be implemented in the form of a neural network. This network can in particular be a “Multi-layer Perceptron” (MLP) type network.

• une couche d'entrée comprenant un nombre de neurones égal au nombre de paramètres d’entrée ;
• une couche ou éventuellement deux couches intermédiaire(s) ; et
• une couche de sortie comprenant un nombre de neurones égal au nombre de classes de sorties.

The following configuration can for example be adopted for the neural network:

• an input layer comprising a number of neurons equal to the number of input parameters;
• a layer or possibly two intermediate layers; And
• an output layer comprising a number of neurons equal to the number of output classes.

For training this neural network, we can prepare a training database as follows.

For a certain number of existing workstations, for each workstation (each being defined by a workstation record, before making the modifications), the list of modifications having been made to the workstation considered is determined.

This list of modifications made constitutes the desired output of the model (‘ground truth data’) for the workstation considered.

Thus in this mode of implementation, the training database includes data associating a record for an existing workstation and one or more modifications having been made to this workstation.

The model is then trained to provide as output the modifications made, for each of the workstations taken into account in the training database. The cost function used for training can be any function to minimize the deviation between the actual model output and the desired model output.

After training, the model thus trained can be used to implement the method below, in accordance with the present disclosure.

Identification and implementation of modification(s) for a set of workstations

For a set of PTi workstations requiring improvement, the preferable modifications are then determined and carried out as follows ( ):

The first step is a data entry step S110, which is carried out for each of the workstations in the same manner as step S10 of the first embodiment.

Then, the following step S120 is carried out for each of the workstations:

S120) we provide as input the record of the workstation to be improved (or part of the data from this record); this recording is then processed by the predictive model: this provides as output a list of modifications considered preferable for the workstation considered PTi.

On the basis of this list, for each of the workstations, the modification(s) identified as preferable in step S120 for the workstation considered are carried out, during an operation S130 similar to operation S50 of the first embodiment.

Then, the update operations are carried out: Operation A for updating the database is carried out in the same way as for the first embodiment.

Operation B, in the second embodiment, consists of updating the predictive model. This update may consist of a new learning session for the predictive model, based on an updated database by taking into account the information added to the database.

Claims (16)

Method for identifying at least one modification to be made to a workstation to be improved, said workstation being designed to accommodate at least one person to carry out a professional activity;
the method comprising the following operations:
transmitting to a computer a workstation record including a plurality of characteristics of the workstation to be improved;
on the basis of the recording of the workstation, using a predictive model implemented by the computer, at least one modification to be made to the workstation to be improved is identified;
the predictive model being configured so as to output said at least one modification to be made to the workstation based on modifications already made to existing workstations
S50) said at least one modification to be made is carried out . Identification method according to claim 1, in which the predictive model is configured so as to output said at least one modification to be made to the workstation based on modifications already made to existing workstations, taking into account at least one individual characteristic of said at least one person carrying out their professional activity at the workstation; and the identification of said at least one modification to be made to the workstation to be improved is carried out by taking into account at least one individual characteristic of said at least one person carrying out their professional activity at the workstation. Identification method according to claim 1 or 2, wherein said modifications already made to existing workstations are modifications made to workstations similar to the workstation to be improved. Identification method according to any one of claims 1 to 3, in which the predictive model comprises a decision tree (AD), and the identification of said at least one modification to be made to the workstation to be improved is carried out by traversing the decision tree (DA). Identification method according to claim 4, according to which during the course of the decision tree, at least one choice is based on a risk variable chosen from a frequency of occurrence (F), a severity (G), a damage (DO) and detectability (DE) of a disaster affecting a person carrying out their professional activity at the workstation. Identification method according to any one of claims 1 to 5, in which the predictive model is configured using a database (BD) comprising workstation records, said records comprising values of risk variables; a value of at least one risk variable being recorded in the form of an element of a Fibonacci sequence, or being associated with a character string. Identification method according to any one of claims 1 to 6, in which the identification of said at least one modification to be made to the workstation to be improved comprises the following operations:
S20) using the predictive model (AD), a list (LPS) of workstations similar to the workstation to be improved is determined;
S40) said at least one modification to be made to the workstation is selected from said list (LPS) of similar workstations. Identification method according to claim 7, in which operation S40) comprises the following operations:
S42) from the list (LPS) of similar workstations, a list (LME) of possible modifications is determined;
S44) among the modifications of the list (LME) of possible modifications, said at least one modification to be made to the workstation is selected. Identification method according to claim 8, in which the operation S44) comprises the following actions:
- for each modification of the list (LME) of possible modifications, a number of times the modification has been made to a workstation in the list (LPS) of similar workstations is calculated;
- said at least one modification to be made to the workstation is selected as being a modification or a set of modifications most or most frequently adopted for the workstation, possibly taking into account the professional activity and/or individual characteristic(s) of said at least one person carrying out their professional activity at the workstation. Identification method according to claim 4 and any one of claims 1 to 9, in which
the decision tree takes into account, as a final variable, a level of risk (NR(PTi)) at the workstations (PTi); And
said at least one modification to be made to the workstation is selected so as to reduce or minimize a level of risk (NR(PTi)) at the workstation to be improved. An identification method according to any one of claims 1 to 3, wherein the predictive model comprises a model trained by machine learning to predict modifications having been made to existing workstations. Identification method according to any one of claims 1 to 3, in which the predictive model comprises a model trained by machine learning to identify said at least one modification to be made to the workstation to be improved so as to reduce a level of risk at the workstation to be improved. Method for identifying modifications to be made to a set of workstations to be improved, in which modifications to be made to the workstations are identified by implementing an identification method according to any one of claims 1 to 12 for each workstation; and, when implementing said method, the following operations are carried out:
for each of the workstations, S20) using the predictive model (AD), a list of workstations similar to the workstation considered is determined; determining said list of similar workstations comprising an operation S26) in which a risk class (y) is determined for the workstation considered, from the risk levels of the workstations of said list of workstations similar; And
S30) an order of processing priority is determined for the workstations, based on the risk classes determined; Then,
operations S40) are carried out to identify said at least one modification to be made, for the workstations, according to the order of priority determined in operation S30). Computer program comprising instructions which, when the program is executed by at least one processor, cause said at least one processor to execute the operations of an identification method according to any one of claims 1 to 13. Non-volatile computer-readable storage medium on which the computer program according to claim 14 is stored. Identification device (10) comprising at least one processor, as well as a storage medium according to claim 15.