BLOG
EYE on Sampling: Optimizing Sampling for Time Efficiency
Introduction
Getting fruit to its perfect ripening stage depends on having an insightful sampling process. Sampling gives us a glimpse into the fruit’s ripeness, allowing us to assess quality and make informed decisions as the final ripening day approaches.
However, simply taking samples isn’t enough, we need to optimize this process to make the most of it. By refining sampling techniques and analysing the results, we can gain deeper insights into ripeness levels, ensuring that the best possible fruit is delivered to the customers.
Practically, optimizing sampling will enhance decision-making in several ways. For example, fruits will be classified more accurately according to the client’s specifications. It will also provide more precise answers to key questions, such as:
– When should fruits be sorted?
– Should the fruits continue ripening, or are they ready for production?
Let’s explore how to fine-tune the sampling strategy and come closer to optimal sampling!
Research Aim
The research goals for sampling optimization depend on the specific needs of the organization. Different companies have unique goals and operational challenges. For some, the focus is on improving the accuracy of their ripeness predictions, as they feel their current methods aren’t precise enough. These organizations are looking for ways to better assess ripeness and ensure consistent quality.
On the other hand, there are companies that are already satisfied with their prediction accuracy but aim to streamline the process by reducing time, waste, and costs. Their goal is to maintain their high standards while operating more efficiently. Our research ensures that the results are tailored to meet the specific needs of each case.
Research Setup
Research was conducted in two distinct environments: purchase order (PO)-based sampling and pallet-based sampling, focusing on avocados during quality control and ripening processes.
– Data Capturing:
The first step involved extensive data capturing. The pressure of sampled avocados was documented using EYE on Fruit, and the fruits were monitored throughout their entire ripening cycle. Based on this pressure data, the avocados were classified into ripening classes that were then used for validation.
– Ripening classes for different methods:
In both pallet-based and PO-based sampling, the ‘true’ ripening classes are determined using a large amount of pressure data from the pallet or PO. This data is either collected directly through EYE of Fruit or generated/linked using it. From the available pressure data, a sample is selected, with the number of sampled fruits adjusted as needed.
The ripening classes for the samples are identified, and further analysis is performed using the organization’s specific characteristics to develop the proposed (smart) ripening classes.
– Error Comparison:
In both environments, we compare the ripening classes of the samples and smart ripening classes to the “true” ripening classes. This allows us to calculate the error for each method. Finally, we compare the errors to determine which approach—sample-based or smart ripening classes—leads to greater accuracy.
Result
PO-based and pallet-based sampling analyses produce similar results. The graphs below clearly demonstrate that the smart ripening classes are more accurate than those derived directly from the samples.
Figure 2 represents the aim of improving the accuracy of ripening class predictions using a sample size of 10, although similar results are seen with other sample sizes. Here, the green bars represent the actual ripening classes, the blue bars the sample ripening classes, and the purple bars the classes derived from our analysis. It’s evident that the smart ripening classes align closer to the actual ripening classes compared to the sample ones. Notably, some hard fruits go undetected in the samples while they can be accurately predicted using our method.
Figure 3 focuses on the aim of maintaining prediction accuracy while reducing waste, time, and costs. The green line represents the error from comparing sample ripening classes to the true ones, while the purple line shows the error from comparing smart ripening classes to the true ones. As illustrated, the error for the smart ripening classes is consistently lower than that of the sample classes across all sample sizes.
Importantly, as the sample size decreases, the performance difference between the two methods (sample and smart) increases, with the smart method significantly outperforming the traditional approach. For example, as it can be seen from the blue line, if an organization has been making decisions based on a sample size of 10 fruits, our proposed method can achieve the same accuracy using only 5 fruits—effectively halving their sample size while still ensuring reliable predictions.
Conclusion
Sampling plays a crucial role in monitoring the ripening process, serving as the key source of knowledge about fruit ripening progression. By using EYE on Fruit and our analysis, organizations can optimize the insights they gain from fruit sampling.
Looking ahead, future research will focus on integrating sampling analysis with non-destructive testing and exploring other fruit types beyond avocados. We invite you to partner with us to create a customized smart sampling strategy that perfectly fits your needs. Let’s elevate your ripening processes together!