Feeding a preform blowing machine

Abstract

The present invention relates to an installation (106) for feeding preforms (102) to a machine (104) for blowing the preforms (102), comprising the following three handling elements: - a bulk conveyor (108) of the preforms (102), - a device (118) for aligning the preforms (102) received from the bulk conveyor (108), and - a guide (130) for moving the aligned preforms (102) as far as the blowing machine (104), in order to feed the blowing machine (104) at a rate (Dc) of loading with preforms (102), at least part of these handling elements having measurable operating characteristics (Dv, Da, Dc) and modifiable operating parameters (Rc, Rr, Rd). The feed installation (106) further comprises: - a system (144, 146, 148) for measuring the operating variables (Dv, Da, Dc); - a device (140) for controlling the handling elements in order to modify the operating parameters (Rc, Rr, Rd) so that the latter follow set values (Rc*, Rr*, Rd*); and - a regulating device (142) designed to implement a model (M) previously trained by machine learning to: receive as input the measured operating variables (Dv, Da, Dc) and a set value (Dc*) for the loading rate (Dc), and supply as output the set values (Rc*, Rr*, Rd*) for the control device (140), so that the loading rate (Dc) complies with the set value (Dc*) for the loading rate (Dc).

Description

FEEDING A PREFORM BLOWING MACHINE Technical field of the invention

[0001] The present invention relates to a preform feeding installation for a preform blowing machine, a method for automatically driving a model of a feeding installation according to the invention, a method for controlling preform handling elements of a feeding installation according to the invention, and a corresponding computer program. Technological background

[0002] A preform feeding system for a preform blowing machine is known from the prior art, of the type comprising the following three handling elements:

[0003] a bulk conveyor for preforms;

[0004] a device for aligning preforms received from the bulk conveyor; and

[0005] a guide for guiding the aligned preforms to the blow molding machine, to feed the blow molding machine at a preform loading rate;

[0006] at least some of these handling elements have measurable operating characteristics and modifiable operating parameters.

[0007] The operating parameters are generally defined by the operator of the feeding system. However, the preform loading rate is usually the limiting factor for preform blowing.

[0008] It may therefore be desirable to plan a power supply installation that makes it possible to overcome at least some of the aforementioned problems and constraints.

[0009] Therefore, a power supply installation of the aforementioned type is proposed, characterized in that it also includes:

[0010] a system for measuring operating variables;

[0011] a control system for handling equipment to modify operating parameters so that the equipment follows instructions; and

[0012] a control system designed to implement a model previously trained by machine learning to: receive as input the measured operating variables and a setpoint for the loading flow rate and provide as output the setpoints for the control system, so that the loading flow rate complies with the setpoint for the loading flow rate.

[0013] Thus, thanks to the invention, it is possible to optimize the loading rate.

[0014] The invention may further include one or more of the following optional features, in any technically feasible combination.

[0015] Optionally, the operating variables include the loading rate.

[0016] Optionally, the bulk conveyor also includes a moving element for bulk preform movement, such as a cleated belt, and the operating parameters include a speed of the moving element for bulk preform movement.

[0017] Optionally, the alignment device also includes at least one moving preform alignment element, such as a rotating plate or two guide rollers, and the operating parameters include a speed of the moving preform alignment element.

[0018] Optionally, the scroll guide also includes at least one moving element for scrolling the preforms, such as a brush, and the operating parameters include a speed for the moving element for scrolling the preforms.

[0019] Optionally, the model is also general, that is, adapted for several types of alignment device and / or several types of scroll guide, the model then being designed to receive as input an identifier of the type of alignment device and / or the type of scroll guide regulated by the regulation device.

[0020] Optionally, the control device is also designed, for each of several types of alignment device and / or scroll guide types, to implement a model dedicated to the type in question.

[0021] Optionally, the model also expects preform characteristics as input, these characteristics including for example one or more of: a preform length, a preform diameter, a material in which the preforms are made.

[0022] A method for training a model of a power supply installation according to the invention is also proposed, comprising:

[0023] the implementation of a preform recirculation system;

[0024] operation of the power supply system with the recirculation device; and

[0025] During operation, reinforcement learning is implemented to adjust the model in order to optimize a reward that is higher the closer the loading rate is to the loading rate setpoint.

[0026] A method for controlling preform handling elements in a preform feeding system for a preform blow molding machine is also proposed; the handling elements include:

[0027] a bulk conveyor for preforms,

[0028] a device for aligning preforms received from the bulk conveyor, and

[0029] a guide for guiding the aligned preforms to the blow molding machine, to feed the blow molding machine at a preform loading rate,

[0030] at least some of these handling elements have measurable operating characteristics and modifiable operating parameters;

[0031] The control process is characterized in that it comprises:

[0032] receiving measurements of operating variables;

[0033] an implementation of a model previously trained by machine learning for:

[0034] receive as input the measured operating variables and a setpoint for the loading flow rate, and

[0035] to provide output instructions for operating parameters,

[0036] so that the loading rate complies with the loading rate instruction; and

[0037] a control of the handling elements to modify the operating parameters so that they follow the instructions.

[0038] Also proposed is a computer program downloadable from a communication network and / or recorded on a computer-readable medium, characterized in that it includes instructions for executing the steps of a process according to the invention, when said program is executed on a computer. Brief description of the figures

[0039] The invention will be better understood with the aid of the following description, given solely by way of example and made with reference to the accompanying drawings in which:

[0040] This is a simplified view of a first preform blowing installation in which the invention is implemented.

[0041] This is a simplified view of a first preform blowing installation in which the invention is implemented.

[0042] This is a block diagram of an automatic drive process for a model used in the blowing installation of the laou de la,

[0043] This is a functional view of any of the blowing installations of the and of the, with a recirculation device,

[0044] This illustrates an example of a general model.

[0045] This illustrates an example of two dedicated models,

[0046] This is a functional view of a computer system that can implement a regulation device and / or a control device, and

[0047] This is a block diagram of a process for controlling handling elements of a preform blowing installation. Detailed description of the invention

[0048] With reference to the, a first preform blowing installation 100 102 in which the invention is implemented, will now be described.

[0049] As is known, each preform 102 has a substantially tubular body with two ends: a closed lower end and an upper end, in a straight (vertical) orientation of the preform 102. The preform 102 also has a neck extending from the upper end of the body. The neck has a flange attached to the upper end of the body and, generally, a thread located above the flange. The preform 102 is preferably made in one piece, for example by injection molding of a material such as polyethylene terephthalate (PET), recycled PET (rPET), or polypropylene (PP). The body extending below the flange is intended to be blown into the shape of a container. The neck is generally unchanged during the blow molding operation, so that, at the preform 102 stage, it has the final shape of the neck of the final container.

[0050] The first blowing installation 100 includes a blowing machine 104 for the preforms 102 and a preform feeding installation 106 for the blowing machine 104. This feeding installation 106 includes various preform handling elements for the preforms 102.

[0051] More specifically, the feeding installation 106 includes first of all a bulk conveyor 108 for the preforms 102. The bulk conveyor 108 thus includes at least one moving element 110 for the bulk movement of the preforms 102. The moving element 110 can be any part involved in the bulk movement of the preforms 102 and is designed to move at a certain speed, denoted Rc.

[0052] For example, as illustrated in Figure 1, the bulk conveyor 108 can be an elevating conveyor comprising, on the one hand, a cleated belt 114 and, on the other hand, a drive roller 116 for the cleated belt 114, this roller 116 being itself driven by a motor 112. The cleated belt 114, the roller 116, and even the motor 112 could be moving elements within the meaning of the invention. Hereafter, the roller 116 will be considered as the moving element 110, so that the speed Rc corresponds to a forward speed of the cleated belt 114. In the case where the roller 116 or the motor 112 were considered as the moving element, the speed Rc could correspond, respectively, to a rotational speed of the roller 116 or to a rotational speed of the motor 112.

[0053] The feeding installation 106 further includes a device 118 for aligning the bulk preforms 102 received from the bulk conveyor 108. The alignment device 118 thus includes at least one moving element 120 for aligning the preforms 102. The moving element 120 can be any part involved in aligning the preforms 102 and is designed to move at a certain speed, denoted Rr.

[0054] For example, as illustrated in the figure, the alignment device 118 is gyroscopic and comprises a rotating plate 124 driven by a motor 122. More specifically, the gyroscopic alignment device has an inlet 126 into which preforms 102 are discharged by the bulk conveyor 108 to a center of the rotating plate 124, near one of its axes of rotation. By gyroscopic effect on the rotating plate 124, the preforms 102 move towards a peripheral edge 128 of the rotating plate 124, where there is a straightening opening (not shown). Each preform arriving at the straightening opening is intended to be straightened by the fact that its body falls by gravity into the opening, while the collar of the preform rests on a periphery of the straightening opening to keep the neck facing upwards. The rotating plate 124 and even the motor 122 could be moving elements within the meaning of the invention.Subsequently, the rotating plate 124 will be taken as the moving element 120, so that the speed Rr corresponds to a rotational speed of the rotating plate 124. In the case where the motor 122 was taken as the moving element, the speed Rr could correspond to a rotational speed of the motor 122.

[0055] The feeding installation 106 also includes a guide 130 for the movement of preforms aligned by the alignment device 118, up to the blowing machine 104.

[0056] The 130 scroll guide, for example, has two rails between which the aligned preforms extend, with their collars resting on these two rails to slide on them.

[0057] When the alignment device 118 is gyroscopic, it is generally located at the same height as the blow molding machine 104, so the preforms 102 need to be pushed to advance along the rails of the guide rail 130. Therefore, the guide rail 130 includes at least one moving element 134 for moving the preforms. The moving element 134 can be any part involved in moving the preforms 102 and is designed to move at a certain speed, denoted Rd.

[0058] For example, the scroll guide 130 is brushed and includes a brush 136, for example a rotating one, for driving the aligned preforms 102, this brush 136 being itself driven by a motor 132. The brush 136 and even the motor 132 could be taken as a moving element 134. Subsequently, the brush 136 will be taken as the moving element 134, so that the speed Rd corresponds to a rotational speed of the brush 136. In the case where the motor 132 were taken as the moving element, the speed Rd could correspond to a rotational speed of the motor 132.

[0059] To regulate the speeds Rc, Rr, Rd, the supply installation 106 also includes a control device 140 for the handling elements 108, 118, 130, so that the speeds Rc, Rr, Rd follow the instructions Rc*, Rr*, Rd*.

[0060] The power supply installation 106 also includes a regulation device 142 designed to implement a model M previously trained by machine learning to receive measured flow rates Dv, Da, Dc as input, in order to provide output setpoints Rc*, Rr*, Rd*.

[0061] Said model M is, for example, a joint model as described in the publication “Mismatched no more: Joint Model-policy optimization for Model-Based RL; Benjamin Eysenbach; Alexander Khazatsky; Sergey Levin; Ruslan Salakhutdinov; 36th Conference on Neural Information Processing Systems (NeurIPS 2022) » https: / arxiv.org / pdf / 2110.02758 or in the publication « Model-based Reinforcement Learning: A survey; Thomas M. Moerland; Joost Broekens; Aske Plaat; Catholijn M. Jonker; Foundations and Trends® in Machine Learning; ISBN: 978-1-63828-057-6; 2023 » https: / www.nowpublishers.com / article / Details / MAL-086. Reinforcement learning involves an autonomous agent (e.g., a robot, chatbot, video game character, etc.) learning which actions to take, based on experience, in order to optimize a quantitative reward over time. The agent is immersed in an environment and makes decisions based on its current state.In return, the environment provides the agent with a reward, which can be positive or negative. Through iterated experiences, the agent seeks optimal decision-making behavior (called strategy or policy, which is a function linking the current state to the action to be executed), in the sense that it maximizes the sum of rewards over time. Reinforcement learning (RL) is based on a Markov decision process (MDP), which provides a framework for the problem of learning to achieve a goal. An agent learns and makes decisions. It reacts to its environment.

[0062] In this case, the input data for training includes, for example, column speed, roller speed, hopper conveyor speed, and the preform flow rate measured in the elevator column, as well as potentially the machine speed and / or the number of inhibited stations and / or the loading mode in the blow molds (e.g., one mold out of two or one mold out of three). This allows for control of the preform flow rate. A continuous flow rate is considered when the preforms move smoothly (including any preform buffer zone). A discontinuous flow rate is considered when there is a shortage of preforms, and a zero flow rate is considered when the preforms are stopped (preform loading has stopped or preforms are blocked) or when there are no preforms (preform storage hopper is empty or there is an upstream blockage).

[0063] This control of the preform flow rate allows for the allocation of "rewards" within the model. Thus, if the feeding is compliant, meaning the preform flow is continuous, as previously discussed, then a "reward" is assigned to the model parameters (for example, the conveyor speed percentage) and to the current action (for example, increasing the hopper belt speed). If the feeding is not compliant, meaning the flow rate is discontinuous, low, or zero, then a "penalty" is assigned to these parameters. It should be noted that, in a hybrid environment, the actions can be discrete and / or continuous. A discrete action might, for example, consist of selecting the conveyor requiring a speed adjustment, and a continuous action might, for example, consist of selecting a conveyor speed (for example, between 7% and 100% for a lifting column).

[0064] In this way, the model M learns to choose the parameter(s) and action(s). These parameters and actions are assigned in the environment (preform feeder), and the resulting "rewards" and "penalties" are processed in the "Reinforce" algorithm of the reinforcement learning model M, which iteratively updates the attached model.

[0065] It should be noted that the attached model as described above can be substituted by any other equivalent reinforcement learning (RL) model, such as an "Actor-Critic" model as described in the publication "O. Dogru et al., "Reinforcement Learning in Process Industries: Review and Perspective," in IEEE / CAA Journal of Automatica Sinica, vol. 11, no. 2, pp. 283-300, February 2024, doi: 10.1109 / JAS.2024.124227." https: / / www.ieee-jas.net / article / doi / 10.1109 / JAS.2024.124227?pageType=en, for example, without departing from the scope of the invention. Thus, the power supply installation 106 comprises several measuring devices 144, 146, 148, together forming a measuring system. More specifically, the feeding installation 106 includes a device 144 for measuring a flow rate Dv of bulk movement of the preforms 102 in the bulk conveyor 108.The feeding installation 106 further includes a device 146 for measuring a flow rate Da of supply of the preforms aligned by the alignment device 118. The feeding installation 106 further includes a device 148 for measuring a flow rate Dc of loading of the preforms 102 into the blow molding machine 104, from the feed guide 130.

[0066] With reference to the, a second preform blowing installation 200 102 in which the invention is implemented, will now be described.

[0067] This second blowing installation 200 is similar to that of the, except for the alignment device 118 and possibly the scroll guide 130.

[0068] Indeed, the alignment device 118 is this time a roller device and thus comprises two rollers 208, called guide rollers, extending one along the other, with an adjustable spacing. The guide rollers 208 are driven in opposite directions by the motor 122. The preforms 102 are thus discharged at one end of the guide rollers 208 by the bulk conveyor 108 and are tossed about by gravity and the rotation of the guide rollers 208 until they assume the upright position between the two guide rollers 208, with their body facing downwards between the two guide rollers 208 and their flange resting on the two guide rollers 208.

[0069] The moving element 110 can thus be one of the guide rollers 208, in which case the speed Rr can be its rotational speed.

[0070] Furthermore, the guide rail 130 is gravity-driven. The rails are inclined so that the aligned preforms 102 slide on them by gravity, without requiring a moving element 134 or a motor 132.

[0071] It will be observed that, in the same way as before, the supply installation 106 also includes a regulation device 142 designed to implement a model M previously trained by machine learning, as described above, to receive measured flow rates Dv, Da, Dc as input, in order to provide output setpoints Rc*, Rr*, Rd*.

[0072] According to one embodiment, with reference to Figures 1 and 2, the device 144 for measuring the flow rate Dv of the bulk movement of preforms 102 in the bulk conveyor 108 may also include a model for preform detection. This model for preform detection may consist, for example, of a YOLOv8 model as described in the publication "YOLO-based Object Detection Models: A Review and its Applications"; Sohan, M., Sai Ram, T., Rami Reddy, CV (2024). A Review on YOLOv8 and Its Advancements. In: Jacob, IJ, Piramuthu, S., Falkowski-Gilski, P. (eds) Data Intelligence and Cognitive Informatics. ICDICI 2023. Algorithms for Intelligent Systems. Springer, Singapore. » https: / link.springer.com / chapter / 10.1007 / 978-981-99-7962-2_39.The model training is done by passing several types of preforms (dimensions, colors, weights) and labeling the captured images, the conversion of the model allowing processing in an artificial intelligence module, such as a Hailo component. TM (Hailo chip) connected to the machine's CPU. Note that image acquisition per section of the elevator column (variant of the) or the flat conveyor belt (variant of the) is synchronized according to the speed of said elevator column or flat conveyor belt, respectively, and the camera's field of view. The number of preforms per column section is transferred to the PLC to calculate a preform throughput per second, which is used in the previously described M model.

[0073] It is quite clear that the said YOLOv8 model can be substituted by any other equivalent model such as an EfficientDet model as described in the publication "EfficientDet: Scalable and Efficient Object Detection; Mingxing Tan, Ruoming Pang, Quoc V. Le; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2020)" https: / doi.org / 10.48550 / arXiv.1911.09070, for example, without going out of the scope of the invention.

[0074] Incidentally, it goes without saying that the device 146 for measuring a flow rate Da of supply of preforms aligned by the alignment device 118 and / or the device 148 for measuring a flow rate Dc of loading of preforms 102 into the blow molding machine 104 may include a model similar to the model of the device 144 for measuring a flow rate Dv of bulk movement of preforms 102 in the bulk conveyor 108, as described above, without going out of the scope of the invention.

[0075] With reference to the, a 300 training method for model M will now be described.

[0076] During a step 302, a preform recirculation device 102 is placed in place of the blowing machine 104. As illustrated in the figure, the recirculation device, bearing the reference 402, is designed to transport the preforms 102 supplied by the scroll guide 130 to the inlet of the bulk conveyor 108.

[0077] Back at the station, the bulk conveyor 108, the alignment device 118, and the scroll guide 130 are then activated (step 304), while the model M is trained using reinforcement learning (step 306). This involves adjusting the model M to optimize a reward that increases as the loading rate Dc approaches the target Dc*. This target Dc* can, for example, be modified during training.

[0078] The process 300 can further include an in-service training, during which the bulk conveyor 108, the alignment device 118 and the scroll guide 130 are operated (step 308) in normal configuration, i.e. with the blowing machine 104 as illustrated in Figures 1 and 2, without the recirculation device 402, while the model M continues to be trained by implementing reinforcement learning (step 310).

[0079] In a first embodiment, the model M is general, meaning it is suitable for several types of power supply systems, for example, for the two power supply installations 106 described in Figures 1 and 2. The model M is thus designed to also receive as input an identifier Id of the type of power supply installation 106 to be regulated. For example, the power supply installation 106 of lapourra can be associated with the value 1 of the identifier Id, and the power supply installation 106 of lapourra can be associated with the value 2 of the identifier Id.

[0080] The model M is then trained successively with each of the types of power supply 106 provided. For example, the model M is first trained using power supply 106 of the, then the result of this first training is trained again using power supply 106 of the.

[0081] With reference to the, in another embodiment, a dedicated model is driven for each type of feed installation 106, as described previously with reference to Figures 3 and 4. Thus, in the example described, a first model M1 is driven for the feed installation 106 of the (i.e. with the gyroscopic alignment device) and a second model M2 is driven for the feed installation 106 of the (i.e. with the roller alignment device).

[0082] Furthermore, whether in the case of a generic model or a dedicated model, the model M can be trained with several types of preforms 102, that is, preforms with different characteristics. In this case, the model M expects as input the characteristics of the preforms 102 used. These characteristics include, for example, one or more of the following: the length of the preforms 102, the diameter of the preforms 102, and the material from which the preforms 102 are made (for example: PET, rPET, or PP).

[0083] On the contrary, the M model can alternatively be trained with a single type of preform 102. In this case, the M model does not expect such characteristics as input.

[0084] With reference to the, the control device 140 and / or the regulation device 142 may be made in the form of a computer system 702 (i.e. a computer) comprising a data processing unit 704 (such as a microprocessor) and a main memory 706 (such as RAM, from the English "Random Access Memory") accessible by the processing unit 704.The computer system 702 further includes, for example, a network interface and / or a computer-readable medium, such as a local medium 708 (like a local hard drive) or a remote medium (like a remote hard drive accessible via the network interface through a communication network) or a removable medium (like a USB flash drive, or a CD, or a DVD, or a Digital Versatile Disc) readable by means of a suitable reader of the computer system 702 (such as a USB port or a CD and / or DVD disc drive). A computer program 710 containing instructions for the processing unit 704 is stored on the local medium 708 and / or downloadable via the network interface.This computer program 710, for example, is intended to be loaded into main memory 706, so that the processing unit 704 can execute its instructions to implement the process of the.

[0085] Alternatively, all or part of the 710 computer program could be implemented as hardware modules, i.e. as an electronic circuit, for example micro-wired, not involving a computer program.

[0086] With reference to the, a method 800 for controlling the handling elements of the or the, will now be described.

[0087] During a step 802, the control device 142 receives measurements of the operating variables Dv, Da, Dc.

[0088] During a step 804, the control device 142 implements model M or one of the models M1, M2, depending on the implementation chosen.

[0089] Thus, the implemented model receives as input the measured operating variables Dv, Da, Dc and a setpoint Dc* of the loading flow rate Dc, and provides as output the setpoints Rc*, Rr*, Rd* of the operating parameters Rc, Rr, Rd, so that the loading flow rate Dc respects the setpoint Dc* of the loading flow rate Dc.

[0090] During a step 806, the control device 140 receives the instructions Rc*, Rr*, Rd* and, during a step 808, commands the handling elements to modify the operating parameters Rc, Rr, Rd so that they follow the instructions Rc*, Rr*, Rd*.

[0091] In conclusion, it is clear that a feeding system such as those described above optimizes the loading rate in the blowing machine.

[0092] It should also be noted that the invention is not limited to the embodiments described above. Indeed, it will be apparent to those skilled in the art that various modifications can be made to the embodiments described above, in light of the information just disclosed to them.

[0093] In the detailed presentation of the invention given above, the terms used shall not be interpreted as limiting the invention to the embodiments set forth in this description, but shall be interpreted to include all equivalents which can be foreseen by a person skilled in the art by applying their general knowledge to the implementation of the teaching which has just been disclosed to them.

Claims

Installation (106) for feeding preforms (102) to a preform blowing machine (104), comprising the following three handling elements: - a bulk conveyor (108) for the preforms (102), - a device (118) for aligning the preforms (102) received from the bulk conveyor (108), and - a guide (130) for moving the aligned preforms (102) to the blowing machine (104), to feed the blowing machine (104) at a preform (102) loading rate (Dc), at least a portion of these handling elements having measurable operating characteristics (Dv, Da, Dc) and modifiable operating parameters (Rc, Rr, Rd); the feeding installation (106) being characterized in that it further comprises: - a system (144, 146, 148) measurement of operating variables (Dv, Da, Dc);- a control device (140) for the handling elements to modify the operating parameters (Rc, Rr, Rd) so that the latter follow setpoints (Rc*, Rr*, Rd*); and - a control device (142) designed to implement a model (M; M1, M2) previously trained by machine learning to: receive as input the measured operating variables (Dv, Da, Dc) and a setpoint (Dc*) for the loading flow rate (Dc) and provide as output the setpoints (Rc*, Rr*, Rd*) for the control device (140), so that the loading flow rate (Dc) complies with the setpoint (Dc*) for the loading flow rate (Dc). Feeding installation (106) according to claim 1, wherein the operating variables (Dv, Da, Dc) include the loading rate (Dc). Feeding installation (106) according to claim 1 or 2, wherein the bulk conveyor (108) comprises a moving element (110) for bulk movement of the preforms (102), such as a cleated belt (114), and wherein the operating parameters (Rc, Rr, Rd) include a speed (Rc) of the moving element (110) for bulk movement of the preforms (102). Feeding installation (106) according to any one of claims 1 to 3, wherein the alignment device (118) comprises at least one movable element (120) for aligning the preforms (102), such as a rotating plate (124) or two guide rollers (208), and wherein the operating parameters (Rc, Rr, Rd) include a speed (Rr) of the movable element (120) for aligning the preforms (102). Feeding installation (106) according to any one of claims 1 to 4, wherein the scroll guide (130) comprises at least one moving element (134) for scrolling the preforms (102), such as a brush (136), and wherein the operating parameters (Rc, Rr, Rd) include a speed (Rd) of the moving element (134) for scrolling the preforms (102). Power supply installation (106) according to any one of claims 1 to 5, wherein the model (M) is general, i.e. adapted for several types of alignment device (118) and / or several types of scroll guide (130), the model (M) then being designed to receive as input an identifier (Id) of the type of alignment device (118) and / or the type of scroll guide (130) regulated by the regulating device (142). Power supply installation (106) according to any one of claims 1 to 5, wherein the control device (142) is designed, for each of several types of alignment device (118) and / or types of scroll guide (130), to implement a model (M1, M2) dedicated to the type considered. Feeding installation (106) according to any one of claims 1 to 7, wherein the model (M) expects input characteristics of the preforms (102), these characteristics including for example one or more of: a length of the preforms (102), a diameter of the preforms (102), a material in which the preforms (102) are made. Method (300) of training the model (M; M1, M2) of a feeding installation (106) according to any one of claims 1 to 8, comprising:- an installation (302) of a device (402) for recirculating the preforms (102);- an operation (304) of the feeding installation (106) with the recirculation device (402); and- during the operation (304), the implementation of reinforcement learning (306) of the model (M; M1, M2) to adjust the latter so as to optimize a reward that is higher the closer the loading flow rate (Dc) is to the setpoint (Dc*) of the loading flow rate (Dc). A method (800) for controlling preform handling elements (102) of a preform feeding installation (106) for a preform blowing machine (102), the handling elements including: a bulk preform conveyor (108), a preform alignment device (118) for preforms received from the bulk conveyor (108), and a guide (130) for moving the aligned preforms (102) to the blowing machine (104), to feed the blowing machine (104) at a preform loading rate (Dc), at least a portion of these handling elements having measurable operating characteristics (Dv, Da, Dc) and modifiable operating parameters (Rc, Rr, Rd); the control method (800) being characterized in that it includes: - a reception (802) of measurements of the operating variables (Dv, Da, Dc); - an implementation (804) of a model (M;M1, M2) previously trained by machine learning to: receive as input the measured operating variables (Dv, Da, Dc) and a setpoint (Dc*) of the loading flow rate (Dc) and provide as output setpoints (Rc*, Rr*, Rd*) of the operating parameters (Rc, Rr, Rd), so that the loading flow rate (Dc) respects the setpoint (Dc*) of the loading flow rate (Dc); and - a control (808) of the handling elements to modify the operating parameters (Rc, Rr, Rd) so that the latter follow the setpoints (Rc*, Rr*, Rd*). Computer program (710) downloadable from a communication network and / or stored on a computer-readable medium, characterized in that it includes instructions for carrying out the steps of a process according to claim 10, when said program is executed on a computer (702).

Images