Construction and Manufacturing

The purpose of predictive maintenance is to reduce or eliminate unplanned downtime due to equipment failure. This is achieved by forecasting when equipment failure is imminent and suggesting actions to minimise the impact of the failure on the manufacturing process. 

Suggested actions may include:

• Corrective actions
• Scheduled maintenance
• The replacement of equipment
• Planned failure.

This leads to higher predictability and reliability in the manufacturing process, increased availability of the systems and significant cost savings. 

The first step in implementing predictive maintenance is to add sensors to the manufacturing systems to monitor and log their operation over time. Machine learning can then be used to find patterns within the logged data and predict possible failures. By including an ability to incorporate new data over time, the model can learn and adapt to better predict failure modes. This allows for better pre-emptive maintenance scheduling on critical equipment.

If you would like to find out the best way to capture and interpret your company’s data, contact AIML to organise a site visit to survey your manufacturing plant. This data will be used to develop machine learning techniques which will enable a predictive maintenance system that meets your needs.

Vision is the primary sense that is used for manipulating objects in the real world. Computer vision allows computers to interpret and understand the visual world; automating the tasks performed by the human visual system. 

Machine vision is used extensively in manufacturing for automatic visual inspection of products, process control and robot guidance. Outputs from machine vision systems are typically pass/fail or object location and orientation (for robot guidance). 

Computer vision differs from machine vision in that it uses machine learning to transform visual images into descriptions of the world and information that can interface with other processes to elicit appropriate decisions or actions. Computer vision can be used in a range of manufacturing activities, including: 

• Identifying product defects 
• Monitoring product and component assembly 
• Reading and tracking barcodes 
• 3D visual inspection.

AIML has extensive experience in the development of computer vision systems and are pioneers in the development of new techniques for computer vision. Let us work with you to understand your specific requirements and develop a system that best meets your needs. Contact AIML for details. 

Machine learning can be applied to develop more accurate and effective demand plans and to optimise supply chains, providing manufacturers with a competitive advantage.

Demand planning is vital in optimising inventory levels, maintaining quality and maximising the speed at which a product reaches the consumer. Accurately forecasting demand requires analysing historical sales data and detecting patterns that impact consumer behaviour (such as the time of year, weather, planned obsolescence, etc). 

Machine learning is ideal for demand planning as it is able to analyse large quantities of data, remove human-operator bias and identify patterns/correlations in the data. Even a small improvement in the accuracy of the demand plan can increase customer service satisfaction, decrease inventory holdings and reduce supply chain costs.

Supply chain management is a series of steps involved in getting a product to the customer. The steps include moving and transforming raw materials into finished products, transporting those products, and distributing them to the end user. There are many variables in the supply chain that can affect the speed and quality of the product that reaches the consumer, including:

• Consumer location
• Manufacturer location
• Supplier(s) location
• Manufacturing throughput
• Transport time and costs
• Warehouse/Distribution Centre size, location, resources, costs
• Order size and frequency
• Service level requirements.

AIML’s work in optimisation techniques can be applied to the problems of demand planning and supply chain optimisation. 

If you want to better understand and define your problem-space, assess the availability of data and plan how to best to obtain the necessary data from your business systems, contact AIML to discuss how machine learning can be used to identify patterns, map and optimise the variables in your data.

Computer Aided Design (CAD)

Computer Aided Design (CAD) is used extensively in manufacturing to precisely represent and visualise objects in three dimensions on a computer. CAD is used to develop 3D models for:

• 3D visualisation
• Analysing nonlinear stress, dynamics and heat transfer
• Assembly planning

The incorporation of machine learning into existing commercial CAD programs is already enabling the:

• Identification of duplicate parts
• Removal of noise during 3D rendering
• Preparation of models for 3D printing
• Provision of material and supplier suggestions. 

Evaluation criteria for designs are often difficult to quantify and generalise and depend on such metrics as targeted accuracy, material availability, manufacturing processes, cost and time. With appropriate training, machine learning algorithms can provide the capability to automatically provide suggestions and evaluate designs within CAD programs. 

AIML’s work in deep learning, 3D modelling and augmented reality can be used to enhance the performance and functionality of CAD software.

Generative Design

In the traditional design process, knowledge and expertise is used to craft products that meet the needs of the end user. The designer needs to understand various principles and processes to adequately generate a final design. Essentially, the end product depends on, and is constrained by, both what the designer knows and ideas that are in their head.

In generative design, a set of goals is input into the generative design software, including design goals, constraints and manufacturing considerations (such as materials, manufacturing methods and cost constraints).

Machine learning algorithms allow the generative design software to make small modifications to the design and evaluate the results, quickly generating design alternatives. Results can then be evaluated, the input parameters modified and the software re-run, quickly converging on an optimal solution. At the same time, the machine learning algorithms in the generative design software continue to learn what works and what doesn’t for use in future design problems.